Complexity-adaptive compression of color images using binary arithmetic coding

1. A method of compressing single color image data stored in memory, the method comprising, for each of a set of pixels of the single color image data:

generating, by a processor coupled to the memory, a prediction error equal to a difference between the single color image data at that pixel and a predicted value of the single color image data at that pixel;

separating the prediction error into multiple factors;

selecting a respective context model for each of the multiple factors into which the prediction error at that pixel has been separated; and

encoding the prediction error based on the context model selected for each of the multiple factors to generate compressed image data.

2. A method as in claim 1 , wherein the multiple factors include (i) a sign of the prediction error at the respective pixel and (ii) an absolute value of the prediction error at that pixel; and

wherein selecting the context model for each of the multiple factors includes:

selecting a first context model for the sign of the prediction error at that pixel; and

selecting a second context model for the absolute value of the prediction error at that pixel.

3. A method as in claim 2 , wherein selecting the first context model for the sign of the prediction error at that pixel includes identifying, as the first context model, a context model of a set of sign context models based on respective signs of prediction errors at a pixel adjacent to and to the west of that pixel and at a pixel adjacent to and to the north of that pixel.

4. A method as in claim 2 , wherein separating the prediction error into multiple factors includes representing the absolute value of the prediction error as a pair including (i) a bit category of the prediction error and (ii) a relative absolute value of the prediction error, the bit category being the most significant bit of the absolute value of the prediction error, the relative absolute value being a difference between (i) the absolute value of the prediction error and (ii) a minimum value of the prediction error over all values of the prediction error having the bit category; and

wherein selecting the second context model for each pixel includes:

selecting a third context model for the bit category of the prediction error at that pixel; and

selecting a fourth context model for the relative absolute value of the prediction error at that pixel.

5. A method as in claim 4 , wherein selecting the third context model for the bit category of the prediction error at each pixel includes:

testing whether the prediction errors at pixels from a group of pixels are all zero to produce a testing result identifying one of a first set of bit category context models and a second set of bit category context models from which the third context model may be selected, the group of pixels containing the pixels adjacent to that pixel to the west, north, northwest, and northeast; and

identifying, as the third context model, a context model of one of the first set of bit category context models and the second set of bit category context models based on the testing result.

6. A method as in claim 4 , wherein selecting the third context model for the bit category of the prediction error at that pixel includes generating a maximum of bit categories at a set of neighboring pixels, the set of neighboring pixels containing the pixels adjacent to that pixel to the west (W), north (N), northwest (NW), and northeast (NE), the set of neighboring pixels further containing the pixel adjacent to W to the west, the pixel adjacent to NW to the west, the pixel adjacent to N to the north, and the pixel adjacent to NW to the north.

7. A method as in claim 6 , wherein selecting the fourth context model for the relative absolute value of the prediction error includes selecting the fourth context model based on i) the third context model for the bit category of the prediction error and ii) the bit category of the prediction error.

8. A method as in claim 4 , wherein the method further comprises:

performing a first binarization operation on the bit category of the prediction error at that pixel to produce a first binarized string; and

performing a second binarization operation on the relative absolute value of the prediction error at that pixel to produce a second binarized string.

9. A method as in claim 8 , wherein performing the first binarization operation on the bit category of the prediction error at each pixel includes computing a unary representation of the bit category of the prediction error at that pixel.

10. A method as in claim 8 , wherein performing the second binarization operation on the relative absolute value of the prediction error includes:

computing a unary representation of the relative absolute value of the prediction error when the bit category of the prediction error is less than or equal to a bit category threshold value; and

computing a uniform binary representation of the relative absolute value of the prediction error when the bit category of the prediction error is greater than the bit category threshold value.

11. A method as in claim 8 , wherein encoding the prediction error based on the context model selected for each of the multiple factors includes:

for a binarized string equal to one of the first binarized string and the second binarized string, for each bit position of the binarized string, forming a binary symbol from the binarized string and a number of previous binarized strings at that bit position;

selecting a binary context model based on the binary symbols and the indices of the first context model, the third context model, and the fourth context model; and

computing an estimated probability distribution for encoding the prediction error based on the binary context model.

12. A method as in claim 11 , wherein forming the binary symbol from the binarized string and a number of previous binarized strings includes, at each bit position, concatenating the binary digits of the binarized string and the previous binarized strings at that bit position.

13. A computerized apparatus constructed and arranged to compress single color image data, the apparatus comprising:

memory for storing the single color image data; and

controlling circuitry coupled to the memory, the controlling circuitry being constructed and arranged to:

generate, by the computerized device and for each pixel of the image, a prediction error equal to a difference between the single color image data at that pixel and a predicted value of the single color image data at that pixel;

separate the prediction error at each pixel into multiple factors;

select a respective context model for each of the multiple factors into which the prediction error at that pixel has been separated; and

encode the prediction error at each pixel based the context model selected for each of the multiple factors to generate compressed image data.

14. An apparatus as in claim 13 , wherein the multiple factors include (i) a sign of the prediction error at the respective pixel and (ii) an absolute value of the prediction error at that pixel; and

wherein the controlling circuitry constructed and arranged to select the context model for each of the multiple factors is further constructed and arranged to:

select a first context model for the sign of the prediction error at that pixel; and

select a second context model for the absolute value of the prediction error at that pixel.

15. An apparatus as in claim 14 , wherein the controlling circuitry constructed and arranged to select the first context model for the sign of the prediction error is further constructed and arranged to identify, as the first context model, a context model of a set of sign context models based on respective signs of prediction errors at a pixel adjacent to and to the west of that pixel and at a pixel adjacent to and to the north of that pixel.

16. An apparatus as in claim 14 , wherein the controlling circuitry constructed and arranged to separate the prediction error into multiple factors is further constructed and arranged to represent the absolute value of the prediction error as a pair including (i) a bit category of the prediction error and (ii) a relative absolute value of the prediction error, the bit category being the most significant bit of the absolute value of the prediction error, the relative absolute value being a difference between (i) the absolute value of the prediction error and (ii) a minimum value of the prediction error over all values of the prediction error having the bit category; and

wherein the controlling circuitry constructed and arranged to select the second context model for the absolute value of the prediction error is further constructed and arranged to:

select a third context model for the bit category of the prediction error at that pixel; and

select a fourth context model for the relative absolute value of the prediction error at that pixel.

17. An apparatus as in claim 16 , wherein the controlling circuitry constructed and arranged to select the third context model for the bit category of the prediction error is further constructed and arranged to:

whether the prediction errors at pixels from a group of pixels are all zero to produce a testing result identifying one of a first set of bit category context models and a second set of bit category context models from which the third context model may be selected, the group of pixels containing the pixels adjacent to that pixel to the west, north, northwest, and northeast; and

identify, as the third context model, a context model of one of the first set of bit category context models and the second set of bit category context models based on the testing result.

18. An apparatus as in claim 16 , wherein the controlling circuitry constructed and arranged to select the third context model for the bit category of the prediction error is further constructed and arranged to generate a maximum of bit categories at a set of neighboring pixels, the set of neighboring pixels containing the pixels adjacent to that pixel to the west (W), north (N), northwest (NW), and northeast (NE), the set of neighboring pixels further containing the pixel adjacent to W to the west, the pixel adjacent to NW to the west, the pixel adjacent to N to the north, and the pixel adjacent to NW to the north.

19. An apparatus as in claim 16 , wherein the controlling circuitry constructed and arranged to select the fourth context model for the relative absolute value of the prediction error is further constructed and arranged to select the index of the fourth context model based on i) the index of the third context model computed for the bit category of the prediction error at that pixel and ii) the bit category of the prediction error at that pixel.

20. A computer program product comprising a non-transitory, computer-readable storage medium which stores executable code, which when executed by a processor, causes the processor to perform a method of compressing single color image data stored in memory, the method comprising:

generating a prediction error equal to a difference between the single color image data at that pixel and a predicted value of the single color image data at that pixel;

separating the prediction error into multiple factors;

selecting a respective context model for each of the multiple factors into which the prediction error at that pixel has been separated; and

encoding the prediction error based on the context model selected for each of the multiple factors to generate compressed image data.