Method and apparatus for motion estimation.

TECHNICAL FIELD

The present invention relates to a method and apparatus for motion estimation, and more particularly relates to a method and apparatus for further a detailed motion estimation, in which uses an error value generated at an estimation of motion for determining a motion vector, in especially DPCM(differential pulse code modulation) method for coding a digital image signal.

BACKGROUND ART

Generally, there are several coding methods for further efficiently compressing transmission data in an image signal processing apparatus using subsequent digital image signals, such as HDTV(high definition television), Digital VTR(digital video tape recorder), and multi-media, etc. The DPCM method of which codes the digital image signal using data correlation which is present between adjacent frames of the image signal.

In the case of coding a difference signal between adjacent frames with the lapse of time using the DPCM method, it is disadvantageous that much more data amount to be coded is required for an active image rather than a still image. However, by seeing a movement location from a certain block of the previous frame to a specific block of the present frame, that is, the best similarity between a specific image of the present frame and a certain image of the previous frame, efficient data compression can be performed because signal difference values of similar blocks between frames are minimum. Accordingly, in the motion estimation by the DPCM system, difference data between a predetermined image block of the previous frame and the corresponding image block of the present frame is coded, thereby an efficiency of transmission to be enhanced. In the motion estimation method, a motion vector represents a moving direction and magnitude of the image block by comparing an image block within the present frame with the corresponding image block of the previous frame.

There are several methods for the motion estimation of a pixel unit. Generally, since a motion between two frames with the lapse of time may have accurately but little integer-pixel unit, errors may happen to a motion compensation, in case using the motion vector of an integer-pixel unit. Therefore, a motion vector of a sub-pixel unit is used for reducing a motion compensation error. A half-pixel unit, as one estimation method of sub-pixel units, is used for the motion estimation. First, the motion estimation method of a half-peel unit in accordance with a conventional method will be briefly described.

The apparatus of FIG. 1 comprises N₁ × N₂ block constructor 1 for receiving an image signal of the present frame and constructing an image block composed of N₁ × N₂ pixels; M₁ × M₂ search range constructor 2 for receiving the image signal of the previous frame and constructing the image block composed of M₁ × M₂ pixels; a first search unit 3 for receiving the image signals output from the N₁ × N₂ block constructor 1 and the M₁ × M₂ search range constructor 2 and detecting a first motion vector MV1 of a pixel unit; a half-pixel interpolating unit 4 for receiving image signals output from the M₁ × M₂ search range constructor 2 and the first search unit 3 and calculating an image signal value in a half-pixel location around reference pixels, using the image block within the previous frame which is determined by the first motion vector MV1 detected from the first search unit 3 and the image blocks which are obtained by moving the image block at intervals of one pixel; a second search unit 5 for receiving the image signals output from the N₁ × N₂ block constructor 1 and the half-pixel interpolating unit 4 and outputting a second motion vector MV2 of a half-pixel unit; and a summing unit 6 for receiving and summing the output signals MV1, MV2 from the first search unit 3 and the second search unit 5, and outputting a motion vector MV.

The N₁ × N₂ block constructor 1 receives the image signal of the present frame and stores the input signal in a size of N₁ × N₂ blocks. Simultaneously with inputting the image signal of the present frame, the M₁ × M₂ search range constructor 2 receives the image signal of the previous frame and constructs a search range of M₁ × M₂ magnitude. The first search unit 3 moves the N₁ × N₂ block of the present frame in a pixel unit within the search range of the previous frame. In comparing data of a pixel unit between two blocks, the first motion vector MV1 of an integer-pixel unit is calculated by a block location within the search range where a mean squared error MSE or a mean absolute error MAE is a minimum. The first motion vector MV1 of an integer-pixel unit generated from the first search unit 3 is output to the summing unit 6 and the half-pixel interpolating unit 4. The half-pixel interpolating unit 4 linearly interpolates the reference pixels of a designated block within the previous frame by the first motion vector MV1 of an integer-pixel unit and the surrounding pixels, calculates half-pixel values, and then outputs the calculated values to the second search unit 5.

In FIG. 2, "○" shows intervals of a pixel unit, and "X" indicates intervals of a half-pixel unit. The second search unit 5 moves the image block of the previous frame corresponding to the first motion vector MV1 of an integer-pixel unit to every direction per half-pixel respectively, searches a location of a minimum motion compensation error among locations X of obtained eight half-pixel motion vectors and the first motion vector MV1, and outputs one of vector components (-¹/₂, 0, ¹/₂), that is, a minutely controlled value of a half-pixel unit. The summing unit 6 sums the first motion vector MV1 of an integer-pixel unit output from the first search unit 3 and the second motion vector MV2 of a half-pixel unit from the second search unit 5, thereby outputting a complete motion vector MV for the motion estimation. For example, the summing unit 6 sums the horizontal component (-¹/₂) of the second motion vector MV2 of a half-pixel unit from the second search unit 5 to the motion vector ( <maths id="math0001" num=""><math display="inline"><mrow><mtext>MV=(x, y)</mtext></mrow></math><img file="EP0579844A1_D0001.tif" /></maths> ) of a pixel unit from the first search unit 3, and then determines the motion vector ( <maths id="math0002" num=""><math display="inline"><mrow><mtext>MV=(x-¹/₂, y)</mtext></mrow></math><img file="EP0579844A1_D0002.tif" /></maths> ).

The above conventional method determines the location of blocks using the motion vector of a pixel interval unit, linearly interpolates corresponding pixel values between the determined block and surrounding pixels of the block. The method uses the interpolated half-pixel values and the block pixel values of the present frame determined by the motion vector of an integer-pixel unit, and selects the most similar block among eight half-pixel motion positions, that is, "X" of FIG. 2. Accordingly, the conventional method has a drawback which takes too long in processing data for estimating the motion vector of a half-pixel unit.

DISCLOSURE OF THE INVENTION

In order to solve the problem, an object of the present invention provides a method for motion estimation to make a compensation of further detailed motion to be possible, by which, in error values used for an estimation of motion vectors between frames, generates error values according to the result of comparing data between predetermined image block within a previous frame determined by the motion vector of a pixel unit and image blocks obtained by moving the image block to every direction per one pixel respectively, compares generated errors, and estimates the motion vector of a sub-pixel unit.

Another object of the present invention provides an apparatus for motion estimation to make a compensation of further detailed motion to be possible, by which, in error values used for an estimation of motion vectors between frames, generates error values according to the result of comparing data between predetermined image blocks within the previous frame determined by the motion vector of a pixel unit and image blocks obtained by moving the image block to every direction per one pixel respectively, compares generated errors, and estimates the motion vector of a sub-pixel unit.

The object of the present invention is accomplished by a method for comparing an image signal between adjacent frames and estimating a motion of the image. The method comprises the steps of constructing a first image block having the image signal of a present frame and a second image block having the image signal of a previous frame; comparing the image signal of a plurality of image blocks which are present within the second image block and have a block size equal to the first image block with the image signal within the first image block, and generating a number of error values; detecting a first motion vector of a pixel unit using the generated error values; comparing the image signals of a reference image block, namely, an image block of the previous frame corresponding to the first motion vector, and surrounding image blocks which are generated by moving the reference image block to every direction per one pixel each with the image signal of the first image block, and outputting error values generated by the result of comparison; generating a second motion vector of a half-pixel unit using the generated error values; and summing the first and second motion vectors.

Another object of the present invention is achieved by an apparatus for comparing an image signal between adjacent frames and estimating a motion of the image. The apparatus comprises means for receiving the image signal of a present frame and constructing a first image block; means for receiving the image signal of a previous frame and constructing a second image block; search means for receiving the image signals output from the first image block constructor and the second image block constructor, detecting a first motion vector of a pixel unit, comparing the image signals of surrounding image blocks which are generated by moving the reference image block within the previous frame corresponding to the detected first motion vector to every direction per one pixel each and the reference image block with the image signal of the first image block, and generating error values obtained by the result of comparison; half-pixel control means for receiving the errors output from the search means and generating a second motion vector of a half-pixel unit; and means for receiving the first and second motion vectors, and adding and outputting them.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 is a block diagram for illustrating a conventional apparatus for motion estimation.

FIG. 2 is a conceptual diagram for showing a pixel unit and a half-pixel unit.

FIG. 3 is a block diagram illustrating an apparatus for the motion estimation in accordance with a preferred embodiment of the present invention.

FIG. 4 is a detailed block diagram illustrating a horizontal half-pixel controller of FIG. 3.

FIGs. 5(a) and 5(b) are conceptual diagrams showing a method for determining motion vectors of a half-pixel unit in the horizontal half-pixel controller.

BEST MODE FOR CARRYING OUT THE INVENTION

A motion vector of a predetermined block in a present frame corresponding to a search range of a previous frame represents as <maths id="math0003" num=""><math display="inline"><mrow><mtext>MV = (x, y)</mtext></mrow></math><img file="EP0579844A1_D0003.tif" /></maths> . Here, the motion vector MV has a horizontal component of motion vector x and a vertical component of motion vector y. Hereinafter, the present invention will be described using a half-pixel unit, a kind of a sub-pixel unit.

FIG. 3 is a block diagram illustrating an apparatus for motion estimation in accordance with a preferred embodiment of the present invention.

The apparatus of FIG. 3 comprises N₁ × N₂ block constructor 31 for receiving an image signal of the present frame and constructing image blocks composed of N₁ × N₂ pixels; M₁ × M₂ search range constructor 32 for receiving the image signal of the previous frame and constructing image blocks composed of M₁ × M₂ pixels; search means for receiving the image signals output from the N₁ × N₂ block constructor 31 and the M₁ × M₂ search range constructor 32, detecting a first motion vector MV1 of a pixel unit, comparing data between five image blocks generated by moving a block of N₁ × N₂ size to every direction per one pixel respectively in the center of a reference pixel within the previous frame corresponding to the detected first motion vector MV1 and the image block within the present frame, and generating error values according to the result of comparison; half-pixel control means 34 for receiving the errors output from the search means 33 and calculating a second motion vector MV2 in a half-pixel interval vertically and horizontally based on the reference pixel; and summing means 37 for receiving the first motion vector MV1 from the search means 33 and the second motion vector MV2 from the half-pixel control means 34, summing and outputting them. The half-pixel control means 34 includes a horizontal half-pixel controller 35 for detecting a horizontal component of the second motion vector MV2 and vertical half-pixel controller 35 for detecting a vertical component of the second motion vector MV2.

In the image signal of two adjacent frames with the lapse of time, the N₁ × N₂ block constructor 31 receives the image signal of the present frame and stores the input signal in a size of N₁ × N₂ block. Simultaneously with inputting the image signal of the present frame, the M₁ × M₂ search range constructor 32 receives the image signal of the previous frame and stores it in a size of M₁ × M₂ block larger than N₁ × N₂ block. The search means 33 receives respectively the image signals of the stored M₁ × M₂ block and N₁ × N₂ block output from the M₁ × M₂ search range constructor 32 and the N₁ × N₂ block constructor 31.

A motion estimation error obtained by estimating the motion vector of an integer-pixel unit is set as P₀. Motion estimation errors which are generated by moving to every direction per one pixel respectively on the basis of the estimated motion vector, are respectively P₁', P₋₁', P₋₁, and P₁. Mean absolute error MAE or mean squared error MSE is used for calculating motion estimation errors. The equations to calculate the horizontal motion estimation errors (P₀, P₁, P₋₁) by the MAE are as follows.<maths id="math0004" num=""><img file="EP0579844A1_D0004.tif" /></maths>    Here, Y represents the image signal of the present frame, Y' shows that of the previous frame, and <maths id="math0005" num=""><math display="inline"><mrow><mtext>N₁=N₂=N</mtext></mrow></math><img file="EP0579844A1_D0005.tif" /></maths> is a block size. The above equations are corresponds to (k, ℓ) times block from the top of the left in frames. The vertical motion estimation errors (P₁', P₋₁') can be calculated by the same method as the above equation.

The search means 33 compares, in a pixel unit, data magnitude between the image block of the present frame and a plurality of image blocks which are present within the search range of the previous frame and have similar sizes, and calculates motion estimation errors by the result of comparison. The first motion vector MV1 of an integer-pixel unit is determined by a block location where the motion estimation error is a minimum, and is output to the summing means 37. The search means 33 calculates the motion estimation errors (P₋₁, P₁, P₀, P₋₁', P₁') by the reference image block within the previous frame determined by the motion vector of an integer-pixel unit and a number of image blocks surrounding the reference image block per one pixel. Then, the calculated motion estimation errors (P₋₁, P₁, P₀, P₋₁', P₁') are output to the half-pixel control means 34. The present invention simultaneously and individually determines the horizontal and vertical components of motion vector of a half-pixel unit. However, since the horizontal half-pixel controller 35 and the vertical half-pixel controller 36 use the same method, described will be only a method for determining the horizontal component of motion vector of a half-pixel unit.

FIG. 4 is a detailed block diagram illustrating the horizontal half-pixel controller of FIG. 3.

The device of FIG. 4 comprises three input terminals 41, 42, 43 for receiving the motion estimation errors (P₁, P₀, P₋₁); a first adder A1 for adding and outputting the input signal P₁ of a first input terminal 41 and the input signal P₀ of a second input terminal 42; a second adder A2 for adding and outputting the input signal P₀ of the second input terminal 42 and the input signal P₋₁ of a third input terminal 43; a first comparator CMP1, coupled a predetermined input terminal to an output terminal of the first adder A1, for comparing and outputting the input signals; a second comparator CMP2, coupled a predetermined input terminal to an output terminal of the second adder A2, for comparing and outputting the input signals; a second multiplier M2 for multipling the output signal of the first adder A1 by a predetermined coefficient and outputting a multiplied value to the second comparator CMP2; a first multiplier M1 for multipling the output signal of the second adder A2 by a predetermined coefficient and outputting a multiplied value to the first comparator CMP1; and a NOR gate NOR for performing a logic NOT OR function for the output signals from the comparators CMP1, CMP2. The device further comprises three output terminals 44, 45, 46 to distinguish each of the output signals from first comparator CMP1, NOR gate NOR and second comparator CMP2.

The motion estimation errors (P₁, P₀, P₋₁) output from the search means 33 are input to the horizontal half-pixel controller 35 of the half-pixel control means 34. Then, the first adder A1 subtracts the input signal P₀ through the second input terminal 42 from the input signal P₁ through the first input terminal 41. The second adder A2 subtracts the input signal P₀ through the second input terminal 42 from the input signal P₋₁ through the third input terminal 43. The first and second multipliers M1, M2 multiply a predetermined coefficient to each of input signals and output the multiplied result to the comparators CMP1,CMP2, connected to the output of the multipliers, respectively. A differece value (a) between P₀ and P₋₁ and that (b) between P₀ and P₁ are respectively input to the second comparator CMP2 and the first comparator CMP1. So, the first comparator CMP1 compares and outputs the output signals from the first adder A1 and the first multiplier M1. The second comparator CMP2 compares and outputs the output signals from the second adder A2 and the second multiplier M2.

FIGs. 5(a) and 5(b) are conceptual diagrams showing a method for determining motion vectors of a half-pixel unit in the horizontal half-pixel controller 35. FIG. 5(a) shows that the motion vector of a half-pixel unit is present at the left from the motion vector MV of a pixel unit, and FIG. 5(b) does that the motion vector MV of a half-pixel unit is present at the right from the motion vector of a pixel unit.

In FIG. 5(a), what P₁ is larger than P₋₁ (that is, b is larger than a) is that the motion vector MV locates about ¹/₂ from the first motion vector MV1 of a pixel unit to the left. In FIG. 5(b), what P₋₁ is larger than P₁ on the basis of P₀ (that is, a is larger than b) is that the motion vector MV locates about ¹/₂ from the first motion vector MV1 of a pixel unit to the right.

In FIG. 5(a), in the case of P₁ is larger than P₋₁ on the basis of P₀, only the output 44 of the first comparator CMP1 is "high". Thus, the horizontal half-pixel controller 35 outputs "-¹/₂", the horizontal component value of the second motion vector MV2. In FIG. 5(b), in the case of P₋₁ is larger than P₁ on the basis of P₀, only the output 46 of the second comparator CMP2 is "high". Thus, the horizontal half-pixel controller 35 outputs "¹/₂", the horizontal component value of the second motion vector MV2. Where P₁ and P₋₁ on the basis of P₀ are equal each other, the output signals of the two comparators CMP1, CMP2 are all "low". Therefore, the NOR gate NOR receives low values and its output signal becomes "high". Accordingly, the horizontal component of the second motion vector MV2 becomes "0".

The summing means 37 receives the second motion vector MV2 and adds it to the first motion vector MV1 of a pixel unit obtained by the search means 33, thereby estimating the second motion vector MV2 of a half-pixel unit.

INDUSTRIAL APPLICATION

A method and apparatus for motion estimation in accordance with the present invention compares image signals of image blocks within a previous frame corresponding to a motion vector detected among error values used for detecting the motion vector of a pixel unit and surrounding image blocks which are generated by moving the reference image block to every direction per one pixel respectively with the image signal of image blocks within a present frame which the motion estimation is performed, and then generates error values according to the comparison results. By vertically comparing the generated error values, a vertical component in the motion vector of a half-pixel unit is detected. By horizontally comparing the generated error values, a horizontal component to the motion vector of a half-pixel unit is detected. Accordingly, the present invention leads to a greater efficiency to reduce the amount of data to be processed for the detection of the motion vector of a half-pixel unit.