Digital pixel image sensor with independent color channel reference signals

TECHNICAL FIELD

P-0001[0001] This invention relates to techniques and apparatus for color image sensing and in particular to improve color balancing.

BACKGROUND OF THE INVENTION

P-0002[0002] Perceived colors depend almost exclusively on surface reflectance, yet the light that reaches the eye is a function of both surface reflectance and scene illumination. White balance actions are used in digital imaging systems to mimic a person's ability to remove the effects of scene illumination from the perception of color.

P-0003[0003] An example of constant color perception in humans is that a white page looks white whether viewed under a yellow Tungsten light or under the blue sky. The human vision system somehow compensates for variations in illumination in order to achieve this result, although the processes through which this is attained are not well understood. A solid-state imaging system should implement a mechanism with a similar effect, otherwise the white page will look light blue under the latter condition and slightly yellow under the former. The goal of the white balance operation is to provide the imaging system with a compensation mechanism that allows it to reproduce constant white colors under different illumination hues.

P-0004[0004] In many imaging systems, much of the balancing process is performed on the analog sensor signal before it is recorded or transmitted. In particular, the white balance process is commonly described by the following two operations:

P-0005[0005] 1. COLOR CHANNEL BALANCING. This operation compensates for the unequal spectral sensitivities of the different color channels in a digital image sensor. The color channels in a typical Bayer pattern color filter mosaic are: Red, Blue, Green-Red, and Green-Blue. The latter two are used to differentiate between green pixels adjacent to red pixels and green pixels adjacent to blue pixels.

P-0006[0006]<b>2</b>. COLOR CONSTANCY (also known as illuminant compensation). This operation attempts to mimic the human visual system's phenomenon of keeping white objects constant. The illumination level in the scene is estimated and then the appropriate color transform is constructed. The estimation is generally carried out in software and fed back to the sensor.

P-0007[0007] White balance operations are usually performed in the analog processing chain of the sensor. They are difficult to implement and may contribute additional noise to the sensor data prior to digital conversion. Because the physical conditions that warrant white balance actions cannot be removed, the only recourse for reducing the hardware/software overhead that they entail, as well as minimizing their noise contributions, is to find alternative approaches.

P-0008[0008] Current image sensors are based on active pixel architectures and provide white balance actions as part of their post-capture analog processing chain. Color balancing operations require four input planes, one for each color channel, and this demands certain management overhead. The estimation of illuminant levels for color constancy operations is usually carried out in software and the linear transformation operation to produce the illuminant compensation is carried out in analog form which limits the dynamic range of the final compensated results.

BRIEF DESCRIPTION OF THE DRAWINGS

P-0009[0009] The features of the invention believed to be novel are set forth with particularity in the appended claims. The invention itself however, both as to organization and method of operation, together with objects and advantages thereof, may be best understood by reference to the following detailed description of the invention, which describes certain exemplary embodiments of the invention, taken in conjunction with the accompanying drawings in which:

P-0010[0010]FIG. 1 is a graph of the relative spectral responses of color channels in an exemplary image sensor.

P-0011[0011]FIG. 2 is a diagrammatic representation of an exemplary Digital Pixel element.

P-0012[0012]FIG. 3 is a diagrammatic representation of an RGGB (Bayer) mosaic pattern for a Pixel Array in accordance with the invention.

P-0013[0013]FIG. 4 is a diagrammatic representation of an RGWB mosaic pattern for a Pixel Array in accordance with the invention.

P-0014[0014]FIG. 5 is a flow chart of one embodiment of the method of the invention.

P-0015[0015]FIG. 6 is a diagrammatic representation of a Digital Pixel Image Sensor according to one embodiment of the invention.

DETAILED DESCRIPTION OF THE INVENTION

P-0016[0016] While this invention is susceptible of embodiment in many different forms, there is shown in the drawings and will herein be described in detail specific embodiments, with the understanding that the present disclosure is to be considered as an example of the principles of the invention and not intended to limit the invention to the specific embodiments shown and described. In the description below, like reference numerals are used to describe the same, similar or corresponding parts in the several views of the drawings.

P-0017[0017] A color image sensor is usually composed of an array of pixel sensors arranged behind a color filter array or mosaic. It may also be constructed using a time-switched color filter or a beam splitter (prism). Often a Bayer pattern is used for the mosaic. Color channel balancing attempts to compensate for the unequal sensitivities of the different color channels in an imaging sensor using a Bayer pattern for its color filter mosaic. FIG. 1 shows the corresponding sensitivities for the Red, Blue, Green-red, and Green-Blue color channels in an exemplary image sensor. The different sensitivities must be compensated for otherwise an acquired image will not have natural color when viewed on a standard color display.

P-0018[0018] The dynamic range of a charge accumulating pixel sensor may be adjusted by varying the reference voltage applied to the pixel. By using the independent triggering mechanism inside pixel sensors with charge accumulation pixels, it is possible to set independent voltage reference signals to each color channel in the Bayer pattern, or any other color filter mosaic, to produce the equivalent effect of color balancing during the capture stage. This approach takes advantage of the parallelism inherent in this sensor to eliminate the color balancing operation—and its noise—from the post-capture analog processing chain. Furthermore, with the ability to control each color channel independently, a green filter location in the Bayer pattern can be replaced with a white—translucent—color filter. This step allows for direct measurement of the illuminant level and provides the ability to make the determination of the proper color transform locally. This eliminates the need for complex feedback mechanisms. Finally, because the voltage reference signals are generated using digital-to-analog converters (DAC) and can therefore have non-monotonic behavior, it is possible to optimize the dynamic range for each color channel independently.

P-0019[0019] The spectral response of red, green, and blue pixels in a Bayer pattern are different because the transmission properties of filters used to create the pattern are different for each color. In other words, the same light incident on different color pixels does not result in the same output value. This effect is compensated for by applying different reference voltage signals to each color channel in the sensor. When using the Bayer pattern, each pixel only captures one color of the RGB triplet needed to fully represent an image. The other two colors needed at each pixel are generated through interpolation algorithms. In the case of an RGBW pattern, the white (W) pixel needs all three colors generated through interpolation.

P-0020[0020] To maintain color constancy, an algorithmic method is used to identify components of a particular color inside an image and so to determine the reference voltage levels that need to be supplied to the sensor in order to keep these components constant even as illumination conditions are changed. These components could be individual pixels or specific regions, the choice is based on the level of complexity of the algorithm and the computational requirements of the application. In most digital imaging systems a single color is generally chosen for this purpose: white. Color constancy, or white color constancy for that matter, is achieved by adjusting the reference voltage signal for each of the sensor's color channels in order to maintain the RGB values of an image's white components the same even as the viewing illuminant is varied. White is generally designated in digital imaging systems as having all three color channel values (RGB) equal to the maximum digital value of the system. In the case of an 8 bit system, R=G=B=255 for a truly white pixel.

P-0021[0021] An example of a plausible pixel-based white balance procedure is as follows:

P-0022[0022] STEP 1: The pixel in row 0, column 0 of the array, a red pixel, records an integration time that is converted via a lookup table to the digital number 216 out of a maximum of 255 for an 8-bit system. Pixel in row 1, column 1, a blue pixel, records an integration time that is converted to the digital value 240. Color channel gains are originally set to unity for all colors.

P-0023[0023] STEP 2: After color interpolation is performed, the color values for pixel (0,0) are: Red=216, Green=209, Blue=235. The values for pixel (1,1) are Red=222, Green=251, Blue=240. The white balance algorithm determines that pixel (1,1) is the closest to true white (Red=255, Green=255, Blue=255) of all the pixels and uses it as reference. It calculates that the sensor blue color channel gain must change to 1.06, the red channel to 1.14, and the green channel to 0.98.

P-0024[0024] STEP 3: Under the same illumination pixel (0,0) is now Red=246, Green=203, Blue=235, pixel (1,1) is now Red=253, Green=248, Blue=255. Pixel (1,1) has been made “whiter” by this procedure. As the illumination changes the gains are varied to make the “whitest” pixel closer to the true white.

P-0025[0025] If the illuminant is known in advance, white balance can be achieved by capturing a reference image of a white surface and adjusting the relative levels of the reference voltages for the color channels so that the reference image is white.

P-0026[0026] Another white balance approach determines the brightest pixels and calculates the average value for each color channel for the bright pixels or for pixel neighboring the bright pixels. The reference voltages are set so that the average values for the each color channel are equal. When white pixels are included in the array, the white pixel values can be used to determine which areas of the image are the brightest.

P-0027[0027] An exemplary time-integrating pixel architecture <b>100</b> is shown in FIG. 2. The pixel architecture <b>100</b> comprises a photo-detector <b>102</b> and a capacitive element <b>104</b> such as a floating diffusion capacitor. In operation, the photo-current charges the floating diffusion capacitor and produces a photo-voltage <b>106</b>. The charge is allowed to move directly from the photo-detector to the diffusion capacitor without the use of a transfer gate. The photo-voltage <b>106</b>, sensed by the floating diffusion capacitor, and a reference voltage <b>126</b> are supplied as inputs to comparator <b>110</b>. The voltage across the capacitor is proportional to the integration time. After an integration time t₁, the photo-voltage <b>106</b> sensed by the floating diffusion capacitor and the reference voltage <b>126</b> are equal. In the pixel architecture shown in FIG. 2, a global counter <b>111</b> is reset at time t=0. The global counter value is supplied to the pixel architecture via pixel data memory decoder <b>112</b> and data bus <b>114</b>. The switch <b>116</b> provides a logic control signal <b>118</b> to selector <b>120</b>, which causes one of the two reference voltages VREFL <b>122</b> and VREFH <b>124</b> to be passed to the output <b>126</b> of the selector <b>120</b>. The reference voltages, VREFL and VREFH, may be provided by a digital-to-analog converter (DAC) <b>140</b>. Initially, the lower of the lower voltage VREFL is selected. When the photo-voltage <b>106</b> exceeds the reference voltage <b>126</b> for a particular pixel, the output <b>128</b> from the comparator <b>110</b> changes value. At that time the global counter value is latched and stored in the first pixel data buffer <b>130</b>. Optionally, the counter value may be negated by passing it through inverter <b>132</b>. Latching the counter values avoids the need to interrogate the comparator output. In the preferred embodiment, the counter value is stored in the pixel data buffer <b>130</b>. One latch is used for each bit in the buffer. In the figure the latching mechanism is assumed to be an integral part of the pixel data buffer in the preferred embodiment. After the counter value has been stored in the first pixel data buffer <b>130</b>, the logic control signal <b>118</b> is switched so that the higher of the two voltages <b>122</b> and <b>124</b> is passed to the output <b>126</b> of the selector <b>120</b>. This provides a higher reference voltage level. When the higher reference voltage level is exceeded, the value of the global counter is latched into the second pixel data buffer <b>134</b>. The first and second pixel data buffers then contain the global counter values corresponding to the times taken for the sense capacitor to reach the two reference voltages. The illumination intensity P_ican therefore be calculated from knowledge of the two reference voltages and the two integration times. The calculation may be performed by an external processor or by a processor integrated with the pixel array. The processor may also perform calculations to determine the reference voltage signals required for white balancing.

P-0028[0028] In FIG. 2, only a single pixel element is shown for ease of illustration. In practice, an array of pixel elements will be integrated in a single device, such as a CMOS device. Pixel data memory decoder <b>112</b> is coupled to a data bus <b>114</b> and is used to decode the data from individual pixel sensors and send it via data bus <b>136</b> to a processor (not shown). Logic elements in switch <b>116</b> are used to route data from the pixel data buffers <b>130</b> and <b>134</b> to the bus <b>114</b>.

P-0029[0029] A D.C. power supply Vdd <b>142</b> is supplied to all of the electronic components in the pixel sensor. This may be 3.3V, for example. When a reset signal is supplied to the gate of transistor <b>144</b>, the transistor turns on and the photo-voltage <b>106</b> reaches a voltage close to the supply voltage Vdd. This resets the pixel sensor. As light falls on the photo-detector, the voltages drops relative to Vdd. The comparator detects when the voltage falls below the level —VREF.

P-0030[0030] In the time integration pixel described above, the pixel comparator triggers a digital capture event when the reference voltage is reached. White balance actions can vary this reference voltage making the time at which the event takes place vary in order to achieve color constancy even as illumination intensities are changing.

P-0031[0031] In an alternative pixel-architecture, voltage measurements and digital conversion are performed by the pixel circuitry. The conversion generally makes use of a comparator to determine the digital representation of the analog signal. The reference voltage to the comparator can be modified according to a decision made by the white balance algorithms. For example, when a ramp signal is used as the reference voltage signal, the slope of the ramp may be varied.

P-0032[0032] In either architecture, the use of independent color channel reference voltages allows more control over the feedback operation than a single reference signal.

P-0033[0033] In a color image sensor, a color filter array may be placed in front of the pixel array, so that different sensors are responsive to light of different wavelengths, as illustrated in FIG. 1. The sensitivities of the pixel sensors are adjusted by varying the reference voltages VREFH and VREFL. In accordance with the present invention, different reference voltages are supplied to different color pixels, allowing the color balance of the array sensor to be adjusted.

P-0034[0034] A section of an imaging array of the present invention is shown in FIG. 3. The pixel sensors <b>304</b>, <b>308</b>, <b>316</b> and <b>320</b> are configured to receive red light, the pixel sensors <b>306</b>, <b>310</b>, <b>314</b> and <b>318</b> are configured to receive green light and the pixel sensor <b>312</b> is configured to receive blue light. The color and amount of light falling on each pixel sensor is determined by the transmission properties of the color filter array (CFA). The pixel sensors <b>304</b>, <b>306</b>, <b>310</b> and <b>312</b> form a Bayer Pattern. A Bayer pattern contains twice as many green pixel sensors as red or blue pixel sensors since the eye is more sensitive to green light. However, other color patterns may be used. The imaging array utilizes a total of five voltage-reference signals to control the color channels: one, VREF0, for the high reference, and four, VREF1 through VREF4, for the low references. The four low references are routed to the individual pixels in a Bayer pattern to allow for different settings in each color channel, while VREF0 is common to all pixels. In one embodiment, the reference signals are generated by independent digital-to-analog converters (DAC) so that it is possible to provide individually controlled, non-monotonic reference voltages to each color channel during the image capture process. In a color system with a RGGB Bayer pattern, VREF1 corresponds to the red pixel, VREF2 and VREF3 correspond to green pixels, and VREF4 corresponds to the blue pixel, as depicted in FIG. 3. The array is arranged so that the (0,0) location is in the upper left corner of the array. The dashed box indicates the color filter mosaic that represents the Bayer pattern.

P-0035[0035] In a further embodiment of the invention, a white pixel is substituted for a green pixel in the color mosaic, creating an RGWB Bayer pattern as shown in FIG. 4. This is possible provided dynamic range of the color channel is sufficient. In this arrangement, white pixel sensors <b>420</b> and <b>404</b> are used instead of green pixel sensors. This color filter pattern allows for direct measurement of the illuminant level for color constancy algorithms. This is possible because the white color channel can be controlled independently to avoid saturation, something that is much harder to do in post-capture manipulations. In STEP 2 of the white balance process described above, there is no need to determine which pixel is closest to white, since white pixels are available. This further simplifies the white balance operation.

P-0036[0036] A flow chart depicting one embodiment of the method of the present invention is shown in FIG. 5. The method is described with reference to a single pixel sensor, but may be applied to each pixel sensor in the array of pixel sensors. Operation begins at start block <b>502</b>. The pixel photo-detector is reset at block <b>504</b>, by discharging the charge accumulation capacitor for example. The lower of the two reference voltages is selected at block <b>505</b>. The output from the photo-detector is integrated at block <b>506</b>. If a floating discharge capacitor is used, the photo-detector output current is integrated by the capacitor to produce a photo-voltage. At decision block <b>508</b> the photo-voltage is compared with the reference voltage. If the photo-voltage is less than the reference voltage as depicted by the positive branch from decision block <b>508</b>, the current value of a global counter (which is supplied to the pixel sensor) is stored into a first pixel data buffer at block <b>510</b>. Flow than continues to decision block <b>512</b>. If the photo-voltage exceeds the reference voltage, as depicted by the negative branch from decision block <b>508</b>, the higher reference voltage is selected at block <b>511</b> and the integration process continues. Decision block <b>508</b> may be implemented as a comparator. Thus the last value stored in the first pixel data buffer will be the counter value at which the photo-voltage reaches the lower reference voltage. The value is thus latched into the pixel data buffer. While the photo-voltage is less than the higher reference voltage as depicted by the positive branch from decision block <b>508</b>, the current value of the global counter is stored into a second pixel data buffer at block <b>510</b>. When the higher reference voltage is exceeded, the current global counter value is no longer written into the second pixel data buffer, so the value when the threshold was crossed is latched into the buffer. At decision block <b>512</b>, a check is made to determine if the maximum time for integration has been exceeded. If the maximum time has been exceeded, as depicted by the positive branch from decision block <b>512</b>, the pixel data buffer is read. If the maximum time has not been exceeded, as depicted by the negative branch from decision block <b>512</b>, the system waits until the time has been exceeded. At block <b>514</b>, the pixel data buffers are read, whether or not the reference voltages have been reached. At block <b>516</b>, the optical power is determined from the counter values read from the pixel data buffers. This allows the relative levels of the different color channels to be determined. In addition, if a RGWB mosaic is used, the illumination level may be determined directly from the power levels of the white pixel sensors. From these levels, the appropriate white balance is determined at block <b>518</b>. At block <b>520</b> the reference voltage levels VREF1 through VREF4 are adjusted to achieve the desired white balance. At decision block <b>522</b>, a check is made to determine if a new image is to be sensed. If a new image is to be sensed, as depicted by the positive branch from decision block <b>522</b>, flow returns to block <b>504</b> and the pixel sensors are reset. If no more images are to be sensed, as depicted by the negative branch from decision block <b>522</b>, the process terminates at block <b>524</b>.

P-0037[0037] One embodiment of a digital pixel image sensor of the present invention is shown in FIG. 6. The digital pixel image sensor <b>600</b> is used in conjunction with an image processor <b>602</b> that may be integrated with the digital pixel image sensor. In this embodiment, the digital pixel image sensor <b>600</b> comprises a sequence generator <b>604</b> that controls a global counter or digital time stamp <b>606</b>. The digital time stamp supplies a timer signal (such as a global counter value) to each element in the pixel array <b>608</b> via a data bus. The pixel array <b>608</b> is composed of a number of pixel elements arranged in a color mosaic pattern and has at least three color channels (red, green and blue). Additional color channels (such as a white channel, or green-blue and green-red channels) may be provided. The sensitivity of each color channel of the image sensor is controlled by reference voltages <b>610</b> supplied by digital-to-analog converter (DAC) <b>612</b>. The DAC, in turn, is controlled by the sequence generator <b>604</b>. The output from each element the pixel array is one or more digital values. The digital values are converted in lookup table <b>614</b> to a digital power value representing the power of the light incident upon the pixel. The digital values, one from each pixel element, are passed to the image processor <b>602</b> via a serial or parallel connection. The image processor may include a color balance module <b>616</b> and a color constancy module <b>618</b>. The processor is configured to calculate the reference voltages that are required to produce the desired color balance. The image processor controls the sequence generator <b>604</b> over feedback link <b>620</b>. The sensitivities of the color channels may be predetermined, so that color balance module <b>616</b> is not required. The color constancy module <b>618</b> may utilize a white color channel to determine the reference voltages that will result in the desired color channel sensitivities. The array output <b>622</b> is obtained from the image sensor and need not be subject to post-capture processing for white-balance, since the white balance has been performed at the capture stage. In video capture, the reference voltage adjustment performed by the image processor need not operate at the capture frame rate, since the illumination does not often change significantly between consecutive frames. This reduces the amount of processing required.

P-0038[0038] A significant competitive advantage is gained, as image sensor arrays become larger and high frame rates are required, by reducing post-capture image processing time and any noise sources associated with it. This effect is achieved by moving the color balancing operations from the post-capture sequence to the image capture stage by taking advantage of the independent pixel referencing in digital pixel sensors and transform the operation from serial to parallel in nature. An advantage in image quality is obtained from the ability to use white pixels to locally calculate color constancy transforms and eliminate the need from complex feedback operations running on software. Finally, because these operations occur during image capture, the stored digital values have inherent information on the dynamic range of each color channel. Optimizing the dynamic range of each color channel, through a non-monotonic reference signal behavior, can enhance the range of applications of future image sensors.

P-0039[0039] Those of ordinary skill in the art will recognize that the present invention has been described in terms of exemplary embodiments based upon use of a digital sampling, time integrating pixel sensor. However, the invention should not be so limited, since the present invention could be implemented using other pixel architectures.

P-0040[0040] Many other variations will also be evident to those of ordinary skill in the art. The embodiment disclosed can be embodied in a variety of systems including digital still and video cameras, optical recording systems and automotive imaging systems.

P-0041[0041] While the invention has been described in conjunction with specific embodiments, it is evident that many alternatives, modifications, permutations and variations will become apparent to those of ordinary skill in the art in light of the foregoing description. Accordingly, it is intended that the present invention embrace all such alternatives, modifications and variations as fall within the scope of the appended claims.