Method and architecture for compressing image data acquired from a bayer color filter array

• US 20060013314A1
• Filed: 06/07/2005
• Published: 01/19/2006
• Est. Priority Date: 06/07/2004
• Status: Active Grant

• Current Assignee: STMicroelectronics SRL (STMicroelectronics NV)
• Original Assignee: STMicroelectronics SRL (STMicroelectronics NV)
• Inventors: Bruna, Arcangelo Ranieri, Buemi, Antonio Vincenzo, Vella, Filippo
• Granted Patent: US 7,676,082 B2
• Time in Patent Office: Days
• US Class Current: 375/240.220
• CPC Class Codes: H04N 19/119 Adaptive subdivision aspects, e.g. subdivision of a picture into rectangular or non-rectangular coding blocks
• CPC Class Codes: H04N 19/124 Quantisation
• CPC Class Codes: H04N 19/136 Incoming video signal characteristics or properties
• CPC Class Codes: H04N 19/17 the unit being an image region, e.g. an object
• CPC Class Codes: H04N 19/176 the region being a block, e.g. a macroblock
• CPC Class Codes: H04N 19/51 Motion estimation or motion compensation
• CPC Class Codes: H04N 19/94 Vector quantisation

Abstract

For each color channel, the process includes gathering Bayer pattern pixel values by pairs, each pair being composed by two successive pixels belonging to the channel along the scanning direction of the pixels of the image, thus each pair of values representing a current input vector, and calculating a predictor vector of the input vector in terms of the differences between the values defining the input vector and a pair of prediction values generated according to a certain criterion, for representing a prediction error. The process further includes quantizing each so calculated predictor vector according to a heavier or lighter degree of quantization depending on whether the predictor vector is representative of an area of relatively uniform color of the image or of an area of relatively abrupt changes of colors of the image, and generating a multibit code representative of the quantized predictor vector of the input vector according to a certain compression ratio.

Claims

1-9. -9. (canceled)

10. A method of compressing image Bayer pattern pixel data to be successively processed to a full color image through color interpolation of each pixel of the full color image, the method comprising:

dividing the Bayer data in three channels of pixel data including a channel for each basic color filtered pixels of the image;

separately compressing each channel data according to a compression process comprising

for each color channel, gathering Bayer pattern pixel values for processing by pairs, each pair including two successive pixels belonging to a same channel along a image pixel scanning direction, each pair of values representing a currently processed input vector,

calculating a predictor vector of the input vector based upon differences between the values defining the input vector and a pair of prediction values representing a prediction error,

quantizing each calculated predictor vector according to one of a heavier and lighter degree of quantization depending on whether the predictor vector is representative of an area of relatively uniform color of the image or an area of relatively abrupt changes of colors of the image, and

generating a multibit code representative of the quantized predictor vector of the input vector based upon a compression ratio.

11. The method of claim 10, wherein quantizing comprises determining a type of image area of which the predictor vector is representative by dividing a vectorial space of the input vector into regions subdivided in an identical number of equal areas, having increasingly larger dimensions as they are located farther from the origin, and by mapping each prediction vector to a target vector of a set of target vectors, according to the region on which the predictor vector falls.

12. The method of claim 10, wherein the image pixel scanning direction comprises a linewise raster mode image pixel scanning direction.

13. The method of claim 10, wherein the pair of prediction values for processing a next pair of gathered pixel values correspond to the values of the current pair of gathered pixel values.

14. The method of claim 10, wherein the pair of prediction values for processing a next pair of gathered pixel values correspond to a second value of the current pair of gathered pixel values.

15. The method of claim 10, wherein only upper quadrants of the input vectorial space are divided in the regions on which to map each predictor vector by recording a sign inversion of the respective prediction errors values for the input prediction vectors falling in third and fourth quadrants, in a one bit flag of the multibit code.

16. The method of claim 10, wherein information of each pair of Bayer pattern pixel data is encoded in the multibit code including:

one bit flag indicating whether the prediction error vector maps in a upper or lower quadrant of the input vectorial space;

five bits identifying the region of division of the upper quadrants of the input vectorial space on which the prediction vector falls;

three bits representing an X-coordinate of the prediction vector; and

three bits representing a Y-coordinate of the prediction vector.

17. A method of decompressing image Bayer pattern pixel data split into three channels of pixel data including a channel for each basic color filtered pixels of the image, the method comprising for each color channel:

evaluating a multibit code representative of a quantized predictor vector of an input vector based upon a compression ratio;

decoding a calculated predictor vector quantized according to one of a heavier and lighter degree of quantization depending on whether the predictor vector is representative of an area of relatively uniform color of the image or an area of relatively abrupt changes of colors of the image; and

reconstructing an input vector based upon addition of the decoded quantized prediction errors and a pair of prediction values representing a prediction error, the input vector indicating a pair of values including two successive pixels belonging to a same channel along an image pixel scanning direction.

18. The method of claim 17, wherein the predictor vector was quantizing by determining a type of image area of which the predictor vector is representative by dividing a vectorial space of the input vector into regions subdivided in an identical number of equal areas, having increasingly larger dimensions as they are located farther from the origin, and by mapping each prediction vector to a target vector of a set of target vectors, according to the region on which the predictor vector falls.

19. The method of claim 17, wherein the image pixel scanning direction comprises a linewise raster mode image pixel scanning direction.

20. The method of claim 17, wherein only upper quadrants of the input vectorial space are divided in the regions on which each predictor vector was mapped by recording a sign inversion of the respective prediction errors values for the input prediction vectors falling in third and fourth quadrants, in a one bit flag of the multibit code.

21. The method of claim 17, wherein information of each pair of Bayer pattern pixel data is encoded in the multibit code including:

one bit flag indicating whether the prediction error vector maps in a upper or lower quadrant of the input vectorial space;

five bits identifying the region of division of the upper quadrants of the input vectorial space on which the prediction vector falls;

three bits representing an X-coordinate of the prediction vector; and

three bits representing a Y-coordinate of the prediction vector.

22. An image processor for compressing image Bayer pattern pixel data to be successively processed to a full color image through color interpolation of each pixel of the full color image, the processor comprising:

an image divider to divide the Bayer data in three channels of pixel data including a channel for each basic color filtered pixels of the image; and

an image compressor to separately compress each channel data according to a compression process comprising

for each color channel, gathering Bayer pattern pixel values for processing by pairs, each pair including two successive pixels belonging to a same channel along a image pixel scanning direction, each pair of values representing a currently processed input vector,

calculating a predictor vector of the input vector based upon differences between the values defining the input vector and a pair of prediction values representing a prediction error,

quantizing each calculated predictor vector according to one of a heavier and lighter degree of quantization depending on whether the predictor vector is representative of an area of relatively uniform color of the image or an area of relatively abrupt changes of colors of the image, and

generating a multibit code representative of the quantized predictor vector of the input vector based upon a compression ratio.

23. The processor of claim 22, wherein quantizing comprises determining a type of image area of which the predictor vector is representative by dividing a vectorial space of the input vector into regions subdivided in an identical number of equal areas, having increasingly larger dimensions as they are located farther from the origin, and by mapping each prediction vector to a target vector of a set of target vectors, according to the region on which the predictor vector falls.

24. The processor of claim 22, wherein the image pixel scanning direction comprises a linewise raster mode image pixel scanning direction.

25. The processor of claim 22, wherein the pair of prediction values for processing a next pair of gathered pixel values correspond to the values of the current pair of gathered pixel values.

26. The processor of claim 22, wherein the pair of prediction values for processing a next pair of gathered pixel values correspond to a second value of the current pair of gathered pixel values.

27. The processor of claim 22, wherein only upper quadrants of the input vectorial space are divided in the regions on which to map each predictor vector by recording a sign inversion of the respective prediction errors values for the input prediction vectors falling in third and fourth quadrants, in a one bit flag of the multibit code.

28. The processor of claim 22, wherein information of each pair of Bayer pattern pixel data is encoded in the multibit code including:

one bit flag indicating whether the prediction error vector maps in a upper or lower quadrant of the input vectorial space;

five bits identifying the region of division of the upper quadrants of the input vectorial space on which the prediction vector falls;

three bits representing an X-coordinate of the prediction vector; and

three bits representing a Y-coordinate of the prediction vector.