Motion detection in video signals

This invention relates to 3D motion adaptive decoding of video signals.

In the National TV Systems Committee (NTSC) standard for television transmission, RGB (Red Green and Blue) signals from a television camera are converted into a luma signal and a chroma signal to allow for efficient transmission. The luma signal, which is typically labeled 'Y', describes the brightness and maximum detail of a picture registered with the television camera. The chroma signal is a combination of two color difference signals R-Y (red minus Y) and B-Y (blue minus Y), both of which are bandwidth limited to about 1 MHz. The respective color difference signals are used to modulate the phase and amplitude of a 3.58 MHz sub-carrier signal. The phase modulation represents the hue, or specific color intended, whereas the amplitude modulation represents the intensity, or saturation, of the color difference signal. The frequency of the sub-carrier is chosen such that the chroma signal can be frequency interleaved with the luma signal to generate a single, composite signal having no energy interference between the chroma and luma signals.

When a television set receives the NTSC composite signal, the composite signal is decoded in a decoder to separate the luma and chroma signals. The chroma signal is then demodulated into R-Y and B-Y component signals that can be used to recreate RGB signals to be displayed on the television screen. Since the luma and chroma signals share the same frequency bandwidth at around 3.58 MHz, and the luma and chroma signals are not prefiltered before they are added, the luma and chroma signals cannot be fully separated in the decoding process. Thus, as a result of the crosstalk between the luma and chroma signals, various decoding artifacts are present. A number of methods have been designed to achieve a better separation of the luma and chroma signals during the decoding process in order to minimize the decoding artifacts.

Fundamentally, there are two methods for separating the luma and chroma signals. Both involve filters that are designed to discern one portion of the composite signal from the other. The first method uses a "notch/bandpass filter," and the second method uses a "comb filter." As will be seen below, there are many different types of comb filters that all have specific advantages and drawbacks.

The notch filter is designed to pass all frequencies of the composite signal, except the frequency of the chroma signal centered at 3.58 MHz. As a result, the chroma signal is removed, but so is also a corresponding portion of the luma signal, resulting in a loss of high frequency luma information. The notch filter is used in parallel with a bandpass filter, which passes only the frequencies in the narrow chroma band and outputs the chroma signal and the high frequency portion of the luma signal. In summary, the notch/bandpass filter method has the advantages of being a simple, low-cost approach with little or no loss of vertical resolution (that is, low frequencies). The drawbacks are that the luma resolution is lost and that heavy display artifacts result when the high frequency luma is treated as chroma (which is known as a "rainbow pattern" artifact), and when chroma is treated as luma (which is known as dot crawls).

The comb filtering methods are based on that the sub-carrier phase is designed to reverse by 180 degrees between two adjacent image lines (and thereby also between two adjacent frames, since there is an odd number of lines within a frame). There are two major types of comb filtering: line combing and frame combing. In both methods, the basic concept involves adding or subtracting two identical signals that have the chroma phase reversed by 180 degrees. When adding the signals, the chroma signal is canceled out and the luma signal is output, and when subtracting the signals, the luma information is lost and the chroma signal is output.

In line combing, one line, two lines or three lines can be used. When a single line is used, the incoming video signal is delayed by one line and the corresponding delayed and un-delayed pixels are added and subtracted, respectively, to obtain the luma and chroma signals. When two lines are used, weighted addition and subtraction of one-line delayed, two-line delayed and un-delayed pixels are used to obtain the luma and chroma signals, respectively. When three lines are used, a correlation is determined between lines 1 and 2, and lines 2 and 3, respectively, and the comb filtering is performed on the two lines that have the best correlation. The three-line adaptive comb filter is often referred to as a 2D adaptive comb filter. All line combing methods produce better results than the notch/bandpass filter method, but still show cross-color, vertical resolution loss, and dot crawls when a single line color is present.

Frame combing is similar to line combing, but instead of comparing two adjacent lines within a single frame, two lines with the same position within two adjacent frames are used for the combing, thereby extending the comb filtering into the temporal domain. However, frame combing can only be performed for still portions of an image. If motion is present, then it is necessary to revert back to 2D-combing. The method of using the motion detection to change comb filtering from line combing to frame combing is referred to as motion-adaptive decoding, or 3D adaptive decoding. The 3D adaptive decoding keeps full horizontal and vertical resolution without dot crawls and cross color for still images. However, motion artifacts, such as large area dot crawls and ghosting shadows, can be present

In the 3D adaptive decoding, the comb filtering is switched between the 3D and 2D comb filtering methods on a pixel-by-pixel basis. If no motion is present, then the 3D comb filter is applied, and if motion is present, then the 2D comb filter is applied. Hence, there is a need to minimize the motion artifacts by accurately detecting whether there is "true motion" between two frames, or whether the perceived motion is "false motion" due to, for example, jitter or noise.

In general, in one aspect, the invention provides methods and apparatus, including computer program products, implementing and using techniques for distinguishing false motion from true motion, in a video signal that is received by a television set. The video signal is received at the television set. The video signal includes a luma component and a chroma component. A motion detection operation is performed for each of the chroma component and the luma component in several frequency regions of the received video signal. Each frequency region has a unique threshold value for distinguishing false motion from true motion, which is dynamically adjusted based on the incoming video signal. This improves the motion detection accuracy and minimizes image artifacts in an image that is described by the video signal and is to be displayed on the television set.

Advantageous implementations can include one or more of the following features. The motion detection operation can be performed on a pixel-by-pixel basis in one or more of: a low frequency luma domain, a high frequency chroma domain, a high frequency luma domain, and each frequency domain where the motion detection operation is performed can have a separate threshold value that is dynamically determined for each pixel that is undergoing the motion detection operation. The threshold for a pixel undergoing the motion detection operation can be determined as a percentage of the difference in pixel values between the pixel undergoing the motion detection operation and one or more neighboring pixels in an image described by the video signal. Performing a motion detection operation can include determining the final motion value to be the maximum chroma motion or luma motion among the chroma and luma motions detected in the different frequency regions. The final motion value can be clipped to be represented by a 5-bit value, and the clipped final motion value can be supplied to a blender that outputs a 2D decoded video signal, a 3D decoded video signal, or a combination of a 2D decoded video signal and a 3D decoded video signal, based on the clipped final motion value.

The low frequency motion for the luma domain can be obtained by performing the following series of operations: subtracting a one-frame delayed pixel from the pixel undergoing the motion detection operation to obtain a motion value, the one-frame delayed pixel and the current pixels having the same positions within two image frames; lowpass-filtering the motion value; converting any negative motion value to a corresponding positive motion value; subtracting a first threshold value determined for the current pixel from the motion value to obtain a low frequency motion value; and if the low frequency motion value after subtraction of the first threshold value is less than zero, setting the low frequency motion value to zero.

The high frequency motion for the chroma domain can be obtained by performing the following series of operations: subtracting a two-frame delayed pixel from the pixel undergoing the motion detection operation to obtain a motion value, the two-frame delayed pixel and the current pixels having the same positions within two image frames; converting any negative motion value to a corresponding positive motion value; subtracting a second threshold value determined for the current pixel from the motion value to obtain a high frequency motion value; if the high frequency motion value after subtraction of the second threshold value is less than zero, setting the high frequency motion value to zero; and lowpass-filtering the high frequency motion value.

The high-frequency motion for chroma and luma with less luma motion can be obtained by performing the following operations: determining whether a one-frame chroma mode or a two-frame chroma mode has been selected; when the two-frame chroma mode has been selected, rectifying and subtracting two one-frame difference signals to obtain a resulting signal; when the one-frame chroma mode has been selected, bandpass filtering, rectifying and subtracting a one-frame delayed pixel from the pixel undergoing the motion detection operation to obtain a resulting signal, the one-frame delayed pixel and the current pixels having the same positions within two image frames; converting any negative motion value of the resulting signal to a corresponding positive motion value; subtracting a third threshold value determined for the current pixel from the motion value to obtain a high frequency motion value; if the high frequency motion value after subtraction of the third threshold value is less than zero, setting the high frequency motion value to zero; and lowpass-filtering the high frequency motion value.

The adjustable threshold value can be determined on a pixel-by-pixel basis based on a detected edge, a detected edge slope, and a sub-carrier detection. A first threshold value can be assigned if a sub-carrier is detected, and a second threshold value can be assigned if no sub-carrier is detected, thereby achieving independent threshold values for chroma and luma portions of an image. The sub-carrier can be detected by searching for temporal and spatial sub-carrier characteristics, including predetermined patterns for the sub-carrier magnitude, the sub-carrier phase, and the sub-carrier sign.

Performing a motion detection operation for each of the chroma component and the luma component in several frequency regions of the received video signal can include: detecting false motion generated for high-frequency chroma still regions of an image by examining one or more of the following characteristics of the received video signal: a spatial chroma signature, a temporal chroma signature, low frequency motion, chroma signal polarity across two or more image frames of the video, and the average value of a sample window across two or more image frames of the video; and generating a mask signal that is free of false motions, based on the examined characteristics of the received video signal. The video can conform to a PAL video standard, a NTSC video standard, or a component video standard. One or more of a synch edge and a color burst region of the video signal can be detected, and a motion variance of the detected synch edge or color burst region can be determined, and the determined variance can be used to set a programmable variable threshold gain percentage for the video signal. The programmable variable threshold gain percentage for the video signal can also be adapted based on the detected chroma and/or the detected luma.

The details of one or more embodiments of the invention are set forth in the accompanying drawings and the description below. Other features, objects, and advantages of the invention will be apparent from the description and drawings, and from the claims.

An embodiment of the invention will now be described in detail, by way of example only, and with reference to the accompanying drawings, in which: <ul id="ul0001" list-style="none"><li>FIG. 1 shows a high-level flowchart of a process performed by a decoder in accordance with the invention.</li><li>FIG. 2 shows a schematic block diagram of a decoder in accordance with the invention.</li><li>FIG. 3 shows a high-level flowchart of the motion detection operation of FIG. 1.</li><li>FIG. 4A shows a schematic block diagram of circuitry for detecting low frequency motion.</li><li>FIG. 4B shows a flowchart of a low frequency motion detection process.</li><li>FIG. 5A shows a schematic block diagram of the logic used to detect high frequency motion for chroma and/or luma.</li><li>FIG. 5B shows a flowchart of a motion detection process for high frequency chroma and/or luma.</li><li>FIG. 6A shows a schematic block diagram of the logic used to detect high frequency motion for chroma and/or luma, with less luma motion.</li><li>FIG. 6B shows a flowchart of a motion detection process for high frequency chroma and/or luma with less motion.</li><li>FIG. 7 shows a schematic block diagram of how NTSC motion is determined.</li><li>FIG. 8 is a schematic block diagram showing the operation of the blender (110) of the decoder in greater detail</li><li>FIG. 9 is a schematic block diagram showing the adaptive threshold logic (120) of the decoder.</li><li>FIG. 10 is a schematic block diagram showing the variable threshold logic block (710) of the adaptive threshold logic (120) in greater detail.</li><li>FIG. 11 shows an example of a 5-tap adaptive threshold window.</li><li>FIG. 12 is a schematic block diagram showing the logic used in the decoder to check the chroma amplitude temporally.</li><li>FIG. 13 is a schematic block diagram showing the logic used in the decoder to do a spatial check of the 4fsc sampling.</li><li>FIG. 14 is a more detailed view of the chroma sample compare block of FIG. 13.</li><li>FIG. 15 shows how a final chroma/luma status value is generated that is used by the variable threshold logic block (710).</li><li>FIG. 16 is a schematic block diagram for detecting low frequency motion for PAL in the decoder.</li><li>FIG. 17 is a schematic block diagram for detecting chroma motion for PAL in the decoder.</li><li>FIG. 18 is a schematic block diagram for detecting luma motion for PAL in the decoder, according to a first implementation of the invention.</li><li>FIG. 19 is a schematic block diagram for detecting luma motion for PAL in the decoder, according to a second implementation of the invention.</li><li>FIG. 20 is a schematic block diagram for the logic that is used to detect PAL motion in the decoder.</li><li>FIG. 21 is a schematic block diagram of the logic used to generate the 3D chroma signal in the decoder.</li><li>FIG. 22 is a schematic block diagram of the logic used to do the final motion determination for a combined NTSC / PAL system.</li><li>FIG. 23 is a schematic block diagram of mask logic used to remove false motions generated for high-frequency colored still regions for PAL signals.</li></ul>

Like reference symbols in the various drawings indicate like elements.

The invention relates to methods and apparatus for detecting motion in video signals, such as composite NTSC, PAL signals or component video signals, which are received by a television set. First, an overview of a decoder implementing the novel motion detection scheme will be given, and then the motion detection in the various frequency regions will be presented. This will be followed by a description of how the individual threshold values are adjusted and a presentation of how the detected motion is used by a blender in the decoder to determine whether to output a 2D or 3D decoded signal. The discussion herein will primarily be focused on NTSC signals, but is equally applicable to PAL signals, or to component video signals, with some minor modifications that will be addressed at the end of this description.

FIG. 1 shows a high-level flowchart of a process (1000) performed by a 3D decoder implementing the novel motion detection scheme. As can be seen in FIG. 1, a composite video signal containing both chroma and luma components is received in the decoder (step 1002). A motion detection operation is then performed on the incoming composite video signal (step 1004). As will be discussed in further detail below, the motion detection operation is performed both for chroma and luma in several frequency regions. Each region has a unique threshold value that is adjustable and therefore allows for better detection of "true motion" versus "false motion," resulting in a decoded image with fewer motion artifacts compared to conventional 3D decoders. The process then examines whether motion was detected or not in the motion detection operation (step 1006). If motion was detected, the process proceeds to step 1008 where 2D logic is used to decode the composite signal. However, if no motion was detected in step 1006, the process continues to step 1010 where 3D logic is used to decode the composite signal.

FIG. 2 shows a schematic block diagram of a decoder (100) for decoding NTSC composite signals received by a television set. The 3D decoder (100) receives a composite video signal (102). The composite video signal (102) is delayed by one and two lines, respectively, in a first and a second line delay (104; 106). A current line signal (L1) along with a one-line delayed signal (L2) and a two-line delayed signal (L3) enters an adaptive 2D decoder (108) with vertical logic, horizontal logic, and single-line logic. The adaptive 2D decoder (108) filters out the chroma signal (2D_C) from the composite video signal, and passes the filtered 2D_C chroma signal to a blender (110), which will be described in further detail below. The one-line delayed signal (L2) is also passed to a delay (112), which compensates for the processing time required for filtering out the chroma signal, to generate the luma signal (Y).

In addition to the adaptive 2D decoder (108), the decoder (100) contains logic for generating a motion-adaptive, 3D chroma signal (3D_C), which is also passed to the blender (110). Two frame memories (114; 116) are used in the detection of chroma motion, as the chroma samples are in phase across two frames (that is, in frames F0 and F2), and a loss-free correlation can easily be determined in the motion detection block (118).

FIG. 3 shows a high-level flowchart of the motion detection operation of FIG. 1 (step 1004), which is performed by the motion detection block (118). As can be seen in FIG. 3, the motion detection process starts by receiving an incoming frame signal, a one-frame delayed signal, and a two-frame delayed signal (step 3002). A first pixel in the incoming frame signal is selected (step 3004). The motion detection block (118) then receives separate threshold values for the low frequency luma domain, the high frequency chroma domain and the high frequency luma/chroma domain, respectively, from the variable threshold logic block (120) (step 3006). As a result of the variable threshold logic block (120) supplying a separate threshold value for each pixel and each frequency domain, the motion detection is significantly improved compared to conventional motion detection techniques. In one implementation, the threshold values are based on the pixel's value as compared to its immediate neighboring pixels. The process determines the maximum motion among the motions in the different frequency domains to be the final motion value for the pixel (step 3008). The process then checks whether all pixels in the incoming frame have been processed (step 3010). If all pixels are done, the process ends. Otherwise, a next pixel in the incoming frame is selected (step 3012), and the process repeats steps 3006 through 3010 until all pixels have been processed. This motion detection process is performed for each received frame.

Returning now to FIG. 2, when the final motion value has been determined, the final motion value is passed from the motion detection block (118) to the blender (110) through a 3-tap median filter (122) and a delay block (124). The median tap filter (122) filters the maximum motion value to remove any sparkle motion or non-motion points in the pixel stream. The 3D chroma signal (3D_C) is generated from a current pixel and a corresponding pixel that is one frame prior, that is, F2-F1, and is then divided by a factor two (126). A band pass filter (128) and a delay (130) are applied to signal before the signal is passed to the blender (110). The band pass filter limits the bandwidth of the 3D chroma signal before the signal reaches the blender (110). In one implementation, the chroma band pass filter (128) is a 5-tap filter with coefficients of -1/4, 0, ½, 0, -1/4, with the middle tap being the current pixel undergoing decoding and two pixels on either side. In another implementation, the chroma band pass filter (128) will be a wide band pass filter with a pass band of around 3 MHz centered around the sub-carrier frequency.

Based on the detected motion in the motion detection block (118), the blender (110) decides whether to output the 2D chroma signal or the 3D chroma signal. If motion is present, the 2D_C signal is output, and if no motion is present, then the 3D_C signal is output from the blender (110). The signal from the blender goes to a color killer (132) that operates to remove chroma information in cases where the proper color cannot be displayed for various reasons, such as for monochrome inputs that lack a color burst signal, and so on. Finally the separated chroma (C) and luma (Y) signals are output to the television set for generation of RGB signals to be displayed on the television screen.

FIG. 4A shows a schematic block diagram of the logic used in one implementation of the motion detection block (118) to detect low frequency motion. FIG. 4B shows a corresponding flowchart for the low frequency motion detection process that is carried out in the motion detection block (118) of FIG. 4A. The low frequency motion is determined by subtracting (202) a one frame delayed sample from an incoming frame, that is, F1-F0 (step 4002). As was discussed above, in NTSC, the sub-carrier for two adjacent frames is 180 degrees out of phase. Since the chroma signal is modulated on the sub-carrier, the result of the subtraction of frames F2 and F0 is that luma is canceled out and chroma and motion is obtained. The signal (C) resulting from the subtraction F2-F1 is lowpass filtered (204) in order to remove the high frequency components including chroma (step 4004). The same signal (C) is also used as input to the logic used to detect high frequency motion for chroma and/or luma, with less luma motion, which will be discussed in more detail below with reference to FIG. 6.

In one implementation, the lowpass filter (204) includes three 5-tap filters (2+2) with coefficients of ¼, 0, ½, 0, ¼, and a 3-tap lowpass filter (1+1) with coefficients of ¼, ½, ¼ at the output. This combination of filters in the lowpass filter (204) helps in reducing noise in the motion path and removes sub-carrier portions that may be incorrectly interpreted as motion. The lowpass filtered signal is then passed to a rectifier (206), which converts negative signals or values to positive signals or values (step 4006), and is then passed to a threshold logic block (208). The threshold logic block (208) subtracts the threshold value determined by the variable threshold logic (120) from the incoming rectified motion value (step 4008). The resulting signal is limited (210), that is, any negative values are clipped to zero while leaving positive values unchanged, to obtain the low frequency motion value (step 4010), and the lowpass filtered motion detection value is output (step 4012).

In one implementation, the third 5-tap filter has a bypass option. The attenuation of the lowpass filter block (204) for NTSC signals can be found in Table 1 below, and the attenuation for PAL signals are listed in Table 2 below. <tables id="tabl0001" num="0001"><table frame="all"><title>Table 1:</title><tgroup cols="2" colsep="1" rowsep="1"><colspec colnum="1" colname="col1" colwidth="78.75mm" /><colspec colnum="2" colname="col2" colwidth="78.75mm" /><thead valign="top"><row><entry namest="col1" nameend="col2" align="center">Attenuation for NTSC signals</entry></row><row><entry namest="col1" nameend="col1" align="center"><b>Frequency</b></entry><entry namest="col2" nameend="col2" align="center"><b>Attenuation</b></entry></row></thead><tbody valign="top"><row><entry namest="col1" nameend="col1" align="center">2 MHz</entry><entry namest="col2" nameend="col2" align="center">-25 dB</entry></row><row><entry namest="col1" nameend="col1" align="center">3 MHz</entry><entry namest="col2" nameend="col2" align="center">-80 dB</entry></row><row><entry namest="col1" nameend="col1" align="center">4 MHz</entry><entry namest="col2" nameend="col2" align="center">-100 dB</entry></row><row><entry namest="col1" nameend="col1" align="center">4.5 MHz</entry><entry namest="col2" nameend="col2" align="center">-60 dB</entry></row><row><entry namest="col1" nameend="col1" align="center">5 MHz</entry><entry namest="col2" nameend="col2" align="center">-40 dB</entry></row><row rowsep="1"><entry namest="col1" nameend="col2" /></row></tbody></tgroup></table></tables><tables id="tabl0002" num="0002"><table frame="all"><title>Table 2:</title><tgroup cols="2" colsep="1" rowsep="1"><colspec colnum="1" colname="col1" colwidth="78.75mm" /><colspec colnum="2" colname="col2" colwidth="78.75mm" /><thead valign="top"><row><entry namest="col1" nameend="col2" align="center">Attenuation for PAL signals</entry></row><row><entry namest="col1" nameend="col1" align="center"><b>Frequency</b></entry><entry namest="col2" nameend="col2" align="center"><b>Attenuation</b></entry></row></thead><tbody valign="top"><row><entry namest="col1" nameend="col1" align="center">2 MHz</entry><entry namest="col2" nameend="col2" align="center">-15 dB</entry></row><row><entry namest="col1" nameend="col1" align="center">3 MHz</entry><entry namest="col2" nameend="col2" align="center">-40 dB</entry></row><row><entry namest="col1" nameend="col1" align="center">4 MHz</entry><entry namest="col2" nameend="col2" align="center">-100 dB</entry></row><row><entry namest="col1" nameend="col1" align="center">5 MHz</entry><entry namest="col2" nameend="col2" align="center">-95 dB</entry></row><row rowsep="1"><entry namest="col1" nameend="col1" align="center">6 MHz</entry><entry namest="col2" nameend="col2" align="center">-46 dB</entry></row></tbody></tgroup></table></tables>

In one implementation, in order to make the switching to 3D softer, the same low frequency motion detection is applied to a one-frame delay and a two-frame delay, respectively, shown in FIG. 4A by blocks 212 through 220. Also here, the signal (D) resulting from the subtraction of signal F1 from F0 is used as input to the logic for detecting high frequency motion for chroma and/or luma, with less luma motion, shown in FIG. 6A. The maximum value of the F2-F motion and the F1-F0 motion, respectively, is determined as the final low frequency luma motion.

FIG. 5A shows a schematic block diagram of one implementation of the logic used in the motion detection block (118) to detect high frequency chroma and/or luma motion. FIG. 5B shows a corresponding flowchart of a motion detection process for high frequency chroma and/or luma. As can be seen in FIGS. 5A and 5B, in order to detect high frequency motion, a two-frame delayed sample is subtracted (302) from an incoming frame, that is, F2-F0 (step 5002). The resulting signal is then rectified (304) (step 5004) and passed to a threshold logic block (306). The threshold logic block (306) subtracts a second threshold value (step 5006) determined by the variable threshold logic (120) from the incoming motion value. The resulting signal is limited (308) (step 5008) and low pass filtered with a 3-tap low pass filter (1+1) with coefficients of ¼, ½, ¼ (step 5010). Since the sub-carrier of F0 and F2 has the same phase, this detection scheme provides proper chroma motion, as well as high frequency luma motion. The resulting signal is then output (step 5012).

In one implementation, the signals F2 and F0 can be bandpass filtered before they are subtracted in step 5002, so that low frequency luma motion is filtered out and only chroma band motion is present. As a result, the resulting signal output in step 5012 will only contain chroma band motion.

FIG. 6A shows a schematic block diagram of the logic used in the motion detection block (118) to detect high frequency motion for chroma and/or luma, with less luma motion. FIG. 6B shows a flowchart of a corresponding motion detection process carried out in the motion detection block (118). As can be seen in FIGs. 6A and 6B, this motion estimation is fairly similar to the high frequency motion estimation discussed above with reference to FIGS. 5A and 5B. The logic shown in FIG. 6A can operate in two modes. In the first mode, the logic operates to detect chroma motion across two frames, with less luma motion. In the second mode, the logic operates to detect one frame chroma motion.

First, the process (6000) determines whether the two-frame chroma mode or the one frame chroma mode should be used (step 6002). In one implementation, this is determined by checking whether a one-frame chroma motion enable bit has been set. If the one-frame chroma motion enable bit has not been set, then the two-frame mode is used and the process continues to step 6004, in which the two one-frame difference signals C (that is, F2-F1) and D (that is F1-F0) from FIG. 4A are rectified. Signal D is then subtracted from signal C (step 6006) and the resulting signal is rectified (step 6014). The threshold logic block then subtracts a third threshold value determined by the variable threshold logic (120) from the incoming rectified motion value (step 6016). The resulting signal is limited (308) (step 6018) and low pass filtered (step 6020) in a 3-tap low pass filter (1+1) with coefficients of ¼, ½, ¼. The resulting signal is then output (step 6022) and the process ends.

If the process determines in step 6002 that the one frame chroma motion enable bit has been set, that is, that one frame chroma motion detection should be used, the process (6000) does not use signals C and D, but instead uses frames F2 and F1. Each of frames F2 and F1 are bandpass filtered (step 6008) and rectified (step 6010). The rectified version of signal F 1 is then subtracted from the rectified version of signal F2 (step 6012), and the resulting difference signal is then rectified (step 6014). After the difference signal has been rectified in step 6014, the process continues with steps 6016 through 6022, as described above.

FIG. 7 shows a schematic block diagram of how the final NTSC motion is determined based on the motions determined for the different frequency regions. The final motion file is the maximum of the motions determined above for the various frequency areas. As was seen in FIG. 2 above, the final motion is then passed through a 3-tap horizontal median filter (122). The 3-tap horizontal median filter (122) removes sparkle motion and no_motion points and/or lines. The final NTSC motion is clipped to be 5 bits wide. A maximum motion value of 16 will cause the decoder to output the 2D luma and chroma signals only. A minimum motion value of zero will cause the decoder to output the 3D luma and chroma signals only. Any values in between zero and 16 will cause the decoder to output a signal that is a blended 2D and 3D signal. This blending is done in the blender (110), which will now be described in greater detail.

FIG. 8 is a schematic block diagram showing the operation of the blender (110) of the decoder in greater detail, in accordance with one implementation of the invention. As was discussed above and can be seen in FIG. 8, the 2D chroma signal (2D_C) comes from the 2D chroma/luma separator, and the 3D chroma signal (3D_C) comes from the 3D chroma/luma separator. The 5-bit motion output from the motion detection block (118) is also used as an input to the blender (110). The chroma output from the blender is based on the 2D_C and 3D_C signals, which are blended as follows:<maths id="math0001" num=""><math display="block"><mrow><mtext>Chroma output = (2D_C * motion)/16 + (3D_C * (16-motion))/16</mtext></mrow></math><img file="EP1549048A2_D0001.tif" /></maths>

The luma output is obtained by subtracting the chroma output from the input composite video (L2). A color killer, which can be enabled and disabled by firmware, is used to remove color from being displayed in cases where the proper color cannot be displayed, for example, when the input composite video (L2) is a monochrome video signal. In some implementations, a filter, such as an adaptive controlled filter, can be applied in the chroma path in order to enhance the cross-color suppression. Similarly, adaptive filters can be put in the luma path in order to reduce cross-luma effects.

As was discussed above, an effective way of decoding the composite signal is to perform motion based adaptive temporal decoding with a 3D comb filter, such that when no motion is present, then temporal combing (3D) is performed, and otherwise spatial combing (2D) is performed. 2D decoding has cross-color and dot crawl artifacts, while 3D decoding does not have these artifacts for still portions of an image. Thus, the motion detection logic is very important in order to accurately detect the motion and thereby achieve a better image quality.

In one implementation, the motion detection is achieved by studying differences between a current frame and previous frames on a pixel-by-pixel basis. The detected difference, that is, the "raw difference" needs to be processed in order to detect true motion. Typically, motion detection logic contains some type of threshold logic, which controls how much raw motion will be suppressed based on system noise. If this threshold value is increased, then the real motion will be lost in some cases, and if the threshold value is decreased, system noise will generate erroneous motion indications. Therefore, it is desirable to have adaptive threshold logic. Furthermore, the more dynamic the adaptive threshold logic is, the better the motion detection logic will perform. In order to understand the various issues that need to be taken into account in the adaptive threshold logic, a brief overview of various noise types will now follow.

The main causes for motion detection inaccuracies are random noise, sampling clock noise or jitter, diagonal transition noise due to camera shimmer, clamp noise and so on. The following discussion will focus on sampling clock noise or jitter, and diagonal transition noise. In addition to random low amplitude noise in video signals, the more high amplitude and high frequency noise for the 3D decoder motion detection logic is edge noise, which causes false motion across frames, and hence switching to 2D in those areas of the image, although 3D should be used. The edge noise occurs as a result of the ADC (Analog to Digital Converter) sampling clock jitter, because of various reasons like clock edge noise, PLL (Phase locked loop) instability, PLL tracking, or the loop filter settings.

As a result of the sampling clock jitter, there are variations in the sampled signal amplitudes at all the signal edges. These variations cause edge noise to be present, which is directly proportional to clock jitter and to the edge amplitude or the slope, that is, the difference in pixel values for two pixels that are located on either side of an edge. In one implementation this difference is determined by determining a first difference of a current pixel and a previous pixel, and a second difference between a current pixel and a next pixel, and selecting the maximum of the two determined differences. In order to address this problem, the threshold needs to be dynamically varied based on both edge detection and on the edge slope, on a pixel-by-pixel basis.

The edge/slope adaptive threshold can be further improved based on sub-carrier detection. Typically in a digitized composite video signal, the edge noise in the high frequency luma portions is higher than the edge noise in the high frequency chroma portions, since the sampling clock is locked to the color burst portion of the sub-carrier signal. Furthermore, the luma edges typically have steeper slopes than the chroma edges, since the luma edges have high resolution up to around 5 MHz. If the edge threshold is set high to mask high frequency luma edge noise, then some associated loss of real chroma motion will also occur. However, if the edge/slope threshold is made adaptive based on sub-carrier detection, the threshold logic can be improved to be more effective, as will now be discussed.

FIG. 9 shows a schematic block diagram of the threshold logic (120) that includes an edge detection logic block (710) and sub-carrier detection (SCD) logic block (720). The edge detection logic block (710), which will be described in further detail below, contains logic that allows a user to set a threshold value as a percentage of the edge, that is, how much of the slope amplitude should be assigned as an edge threshold.

This edge-based threshold is then made further adaptive through the SCD logic block (720), which examines the sub-carrier status. The SCD logic block (720) will also be described in further detail below, but generally the SCD logic block (720) operates and interacts with the edge detection logic block (710) as follows. A user may set two percentages for a particular edge, for example, x% and y%. If the SCD logic block (720) does not detect a sub-carrier, then x% is used as the slope percentage threshold value, and if the SCD logic block (720) detects a sub-carrier, then y% is used as the slope percentage threshold value for that particular edge. If x% is set to be higher than y%, a higher slope percentage threshold value can be used for luma portions of an image and a lower slope percentage threshold value can be used for chroma portions of the image, thereby reducing the chroma motion loss and achieving a dynamic threshold variation based on SCD on a pixel by pixel basis. Because of this independent control for luma and chroma portions, the threshold logic can be programmed to suppress significant luma edge noise without much loss in chroma motion.

The SCD detection block (720) detects the sub-carrier by checking for sub-carrier characteristics, both spatially and temporally. In the spatial domain, because the sampling clock is four times the sub-carrier frequency and locked to the sub-carrier through the color burst, the alternate sub-carrier samples will be nearly equal in magnitude, but opposite in phase or sign. Furthermore, the sub-carrier changes phase by 180 degrees across lines, which can also be used. In the temporal domain, the sub-carrier inverses phase by 180 degrees every frame, so there will be a significant difference between two adjacent frames and much less difference between frames that are two frames apart. Thus, by studying both the spatial domain and the temporal domain, an approximate sub-carrier detection can be made to control the threshold logic. As can be seen in FIG. 9, the SCD block (720) receives the three frame taps F2, F1, and F0 (that is, the current frame, the previous frame, and the frame prior to the previous frame) for sub-carrier detection.

FIG. 10 shows a schematic block diagram of the variable threshold logic block (710). As can be seen in FIG. 10, the composite video signal comes in, and a current pixel (Pn) is selected, for which variable thresholds are to be determined. The logic compares the current pixel (Pn) with a previous pixel (Pn-1) and a next pixel (Pn+1) in the horizontal direction, and selects the maximum difference, as was described above. This difference describes the detected maximum slope amplitude. A corresponding determination is also made in the vertical direction in the Vertical Difference Evaluation block by selecting the current pixel (Pn) for a current line, and the pixel in the same position in a previous line and a next line, respectively. Once the maximum horizontal and vertical slopes, respectively, have been determined, a set of threshold values for the current pixel is generated. The threshold values are linearly proportional to the detected slope value. The proportionality constant can be varied. In some implementations the threshold values can be non-linear with respect to the detected slope value. Also, minimum and maximum clipping functions can be programmed to act on the calculated thresholds so that the final threshold values can be programmed to have some minimum thresholds or be programmed not to exceed some upper limit. In one implementation, the variable threshold logic block (710) can determine threshold values for several adjacent frames, such as F2, F1 and F0 and determine the final threshold value to be the maximum or minimum value determined for the respective frames.

The variable threshold logic helps in removing false motion because of pixel value variations at signal edges due to sampling clock jitter / variations. The higher the signal edge, the larger the variation will occur for the same amount of jitter. Therefore, the pixels that are located on a steeper slope should have a higher threshold value. At the same time, at these edges the motion value is high, so the threshold will not mask the real motion. The variable threshold is generated by taking a percentage, for example, 10 % of the slope that the pixel is located on. This percentage is further adaptively determined based on the color burst variance measures (that is, samples that are two frames apart).

In one implementation, a 3-tap horizontal window is combined with a 3-tap vertical window to form a 5-tap adaptive threshold window. FIG. 11 shows an example of such a 5-tap adaptive threshold window. The current pixel under threshold determination is located in the middle of the respective windows. The absolute difference of the current pixel and the adjacent pixels are determined and the maximum difference among the determined differences is determined to be the slope. For example, in FIG. 11, the current pixel value Pn is compared with the values 1, 2, 3 and 4, respectively, which are the values of the surrounding pixels, and the maximum difference is taken as the final slope from which the threshold is derived for the pixel Pn. The percentage value is made adaptive based on luma/chroma pixel values, for example, x% or y% based on chroma/luma detection.

In one implementation a horizontal threshold value is computed in accordance with the following expression:<maths id="math0002" num=""><math display="block"><mrow><mtext>Vthr(Horizontal) = [[Max{|Pn-(Pn-1)|, |Pn-(Pn+1)|}] * (Programmable attenuation)] +</mtext><mspace linebreak="newline" /><mtext> Programmable DC threshold (min level)</mtext></mrow></math><img file="EP1549048A2_D0002.tif" /></maths>

In one implementation, the decoder can detect a synch edge or color burst regions and determine their motion variance, for example, by using firmware. This detected variance, which can be for one frame motion or two frame motion, can be used to program the linear percentage threshold value, that is the programmable threshold gain. The programmable variable threshold percentage gain can, additionally, be adapted based on detected chroma/luma. The threshold logic also has a programmable minimum DC offset. The max clipping logic clips the output level to the upper limit value if the threshold crosses the programmed upper limit value. The same logic has also been put in the vertical domain, i.e., across lines. As can be realized by one skilled in the art, the threshold percentage and the DC offset for horizontal and vertical taps can be separately programmed. Hence, since line memories are required for vertical logic, only the horizontal logic can be kept, unless the 2D-logic line memories can be shared. Finally, as can be seen in FIG. 9, the variable threshold value is subtracted from the motion value in the motion detection logic block (730) to generate a motion_out value.

The chroma / luma detect logic block (810) of FIG. 10 coarsely detects the chroma / luma signals by checking for chroma characteristics/signature based on two factors. The first factor is based on 4fsc sampling. If the pixel under investigation is part of a chroma cycle, then every alternate sample should be nearly equal in magnitude, but be opposite in sign. After rectification, the absolute difference of adjacent pixels among the four chroma pixels should be nearly the same. The second factor is a chroma amplitude check. If the one-frame difference is greater than the two-frame difference for still images, then chroma is detected, and otherwise the pixel is not a chroma pixel.

Based on these two checks, it can be determined whether the pixel under decoding is on the chroma cycle or on the luma signal. Further, based on this determination, the variable threshold percentage is decided differently for chroma/luma pixels, for example, as 30% for luma and 10% for chroma. It should be noted that the threshold percentage (gain) and the DC offset min and max clip values are independently programmable for each motion path.

A more detailed view of the different parts of the chroma / luma detect logic (810) is shown in FIGS. 12-15. FIG. 12 shows the logic for the temporal chroma amplitude check, which checks the one-frame and two-frame differences, respectively, and outputs a first chroma/luma status value of one or zero, where a value of one represents chroma.

FIG. 13 shows the logic for the 4fsc spatial sampling check, which checks whether every alternate sample is nearly equal in magnitude, but opposite in sign, and outputs a second chroma/luma status value of one or zero, where a value of one represents chroma. FIG. 14 shows a more detailed view of the chroma sample compare block of FIG. 13, where all the four flip-flops are set to one if chroma is detected. Finally, FIG. 15 shows how the first and second chroma/luma status values are compared to generate a final chroma/luma status value that is used by the variable threshold logic block (710).

PAL (Phase Alternating Line) is a video standard used, for example, in Europe, and differs from the NTSC standard in a number of ways. For the purpose of this present application, the most significant difference is that the sub-carrier is out of phase by 180 degrees across two frames, and not across one as in NTSC. Consequently, the sub-carrier is in phase every four frames. For PAL, the 3D chroma signal is generated by subtracting video signals that are two frames apart. The chroma motion is detected by using current, one-frame and two-frame delayed signals (i.e., the F2, F1, and F0 taps).

FIG. 16 shows a schematic block diagram for detecting low frequency motion for PAL and is analogous to the detection of low frequency motion for NTSC, which was described above with respect to FIGS. 4A and 4B, except that for PAL only a two-frame difference is used.

FIG. 17 shows a schematic block diagram for detecting chroma motion for PAL, and is analogous to the detection of chroma motion for NTSC, which was described above with respect to FIGS. 5A and 5B, except that for PAL the F0 and F2 signals are added and clipped, instead of subtracted.

FIG. 18 shows a schematic block diagram for detecting M5_P luma motion for PAL only. It should be noted that the logic depicted in FIG. 18 only gives proper motion for luma only regions. If the logic of FIG. 18 is used on colored regions, large false motions will result. The M5_P motion will be explained in further detail below.

FIG. 19 shows a schematic block diagram for detecting M6_P luma motion for PAL only. It should be noted that the logic depicted in FIG. 19 only gives proper motion for luma only regions. If the logic of FIG. 19 is used on colored regions, large false motions will result. The M6_P motion will be explained in further detail below.

FIG. 20 shows a schematic block diagram for detecting PAL motion. The PAL1 signal, which is the maximum of signals M5_P, M6_P, and M3_NP obtained as shown in FIGS. 18, 19 and 5A, respectively, is used to detect the full range of luma motion. The PAL2 signal, which is the maximum of signals M1_NP, M2_NP, M3_P, M4_NP, and M8_P obtained as shown in FIGS. 4A, 16, 6A and 17, respectively, is used to detect chroma motion and low frequency luma motion. The PAL motion is determined to be the minimum of PAL 1 and PAL2.

As can be seen in FIG. 20, the M8_P and M4_P PAL motion paths have mask logic (M8 Mask) that eliminates false motion generated for high frequency colored still regions. The M8 Mask logic is required because the chroma signal does not cancel by two-frame addition for high frequency colored regions. The chroma signal contains luma near the sub-carrier frequency, which causes the video signal to be not exactly 180 degrees out of phase across two frames and results in false motion. Thus, the M8 Mask logic masks false motion for still high frequency colored regions, and also false motion due to chroma color transitions. A more detailed view of the M8 Mask logic can be seen in FIG. 23. As can be seen in FIG. 23, the M8 Mask logic receives the following inputs: M1_NP, M2_NP, M3_P, M4_NP, M5_P, M6_P, M8_P, F2, F1 and F0, CLD STATUS, M8 Mask En, and M8_Mask_threshold_control_regs. The M8 Mask logic examines signal characteristics, such as spatial chroma signature, temporal chroma signature, low frequency motion, chroma signal polarity across F2, F1, and F0, and the average value of a 4-sample window across F2, F1 and F0, and generates a mask signal (M8_Mask). In the implementation shown in FIG. 23, stretch/expansion logic is also provided for the M8_Mask signal before the mask signal is further processed by the PAL motion detection block.

In one implementation, there is also a small mask logic provided for the NTSC modes, when one frame chroma is used. When one frame chroma motion is used, false motions may be generated by the one frame chroma motion circuit due to the way the composite signal is formed. The mask logic detects and masks these false motions.

FIG. 21 shows a schematic block diagram of how the 3D chroma signal is generated. As can be seen in FIG. 21, for PAL the 3D chroma generation is done using frames F2 and F0, whereas for NTSC, the 3D chroma generation is done using frames F2 and F1. There is also a bypass option for the 3D chroma band pass filter. In one implementation, the default is to use the bypass option, so that we full bandwidth is used for the chroma signal. However, if for some reasons the large bandwidth creates artifacts, such as chroma overshoots, for example, the low frequency motion can be disabled by enable the band pass filter on this path.

Finally, FIG. 22 shows a schematic block diagram of the final motion determination for a combined NTSC / PAL system. The NTSC motion is the maximum of the M1_NP, M2_NP, M3_NP and M4_NP signals, as described above with respect to FIG. 7. The PAL motion signal is generated as described above with respect to FIG. 20. Based on an NTSC/PAL mode detection signal either the NTSC_MOTION or the PAL_MOTION signals is selected. The selected motion signal is then filtered in a 3-tap median filter to remove spikes, and is finally clipped to 5 bits such that the motion signal has a maximum value of 16. The motion signal is then input to the blender (110) of FIG. 2, which combines the 2D and 3D decoding as described above.

The invention can be implemented in digital electronic circuitry, or in computer hardware, firmware, software, or in combinations of them. Apparatus of the invention can be implemented in a computer program product tangibly embodied in a machine-readable storage device for execution by a programmable processor; and method steps of the invention can be performed by a programmable processor executing a program of instructions to perform functions of the invention by operating on input data and generating output. The invention. can be implemented advantageously in one or more computer programs that are executable on a programmable system including at least one programmable processor coupled to receive data and instructions from, and to transmit data and instructions to, a data storage system, at least one input device, and at least one output device. Each computer program can be implemented in a high-level procedural or object-oriented programming language, or in assembly or machine language if desired; and in any case, the language can be a compiled or interpreted language. Suitable processors include, by way of example, both general and special purpose microprocessors. Generally, a processor will receive instructions and data from a read-only memory and/or a random access memory. Generally, a computer will include one or more mass storage devices for storing data files; such devices include magnetic disks, such as internal hard disks and removable disks; magneto-optical disks; and optical disks. Storage devices suitable for tangibly embodying computer program instructions and data include all forms of non-volatile memory, including by way of example semiconductor memory devices, such as EPROM, EEPROM, and flash memory devices; magnetic disks such as internal hard disks and removable disks; magneto-optical disks; and CD-ROM disks. Any of the foregoing can be supplemented by, or incorporated in, ASICs (application-specific integrated circuits).

To provide for interaction with a user, the invention can be implemented on a computer system having a display device such as a monitor or LCD screen for displaying information to the user. The user can provide input to the computer system through various input devices such as a keyboard and a pointing device, such as a mouse, a trackball, a microphone, a touch-sensitive display, a transducer card reader, a magnetic or paper tape reader, a tablet, a stylus, a voice or handwriting recognizer, or any other well-known input device such as, of course, other computers. The computer system can be programmed to provide a graphical user interface through which computer programs interact with users.

Finally, the processor optionally can be coupled to a computer or telecommunications network, for example, an Internet network, or an intranet network, using a network connection, through which the processor can receive information from the network, or might output information to the network in the course of performing the above-described method steps. Such information, which is often represented as a sequence of instructions to be executed using the processor, may be received from and outputted to the network, for example, in the form of a computer data signal embodied in a carrier wave. The above-described devices and materials will be familiar to those of skill in the computer hardware and software arts.

It should be noted that the present invention employs various computer-implemented operations involving data stored in computer systems. These operations include, but are not limited to, those requiring physical manipulation of physical quantities. Usually, though not necessarily, these quantities take the form of electrical or magnetic signals capable of being stored, transferred, combined, compared, and otherwise manipulated. The operations described herein that form part of the invention are useful machine operations. The manipulations performed are often referred to in terms, such as, producing, identifying, running, determining, comparing, executing, downloading, or detecting. It is sometimes convenient, principally for reasons of common usage, to refer to these electrical or magnetic signals as bits, values, elements, variables, characters, data, or the like. It should remembered however, that all of these and similar terms are to be associated with the appropriate physical quantities and are merely convenient labels applied to these quantities.

The present invention also relates to a device, system or apparatus for performing the aforementioned operations. The system may be specially constructed for the required purposes, or it may be a general-purpose computer selectively activated or configured by a computer program stored in the computer. The processes presented above are not inherently related to any particular computer or other computing apparatus. In particular, various general-purpose computers may be used with programs written in accordance with the teachings herein, or, alternatively, it may be more convenient to construct a more specialized computer system to perform the required operations.

A number of implementations of the invention have been described. Nevertheless, it will be understood that various modifications may be made without departing from the scope of the invention. For example, the invention has been described above with reference to NTSC and PAL video signals, which are typically used in broadcasting TV signals, but the invention is also equally applicable to component video, such as a video signal that is output from a DVD player. Accordingly, other embodiments are within the scope of the following claims.