Lossy compression of high-dynamic range image files

1. A computer-enabled method of lossy compressing a high dynamic range (HDR) image, the method comprising:

mapping floating point image data values of the HDR image from a light linear space to a perceptually linear space using a first non-linear mapping function, wherein a value in the light linear space is proportionate to the power of a light source, and wherein a value in the perceptually linear space is proportionate to a visual perceived brightness;

determining floating point discrete cosine transform (DCT) frequency coefficients of the floating point image data values of the HDR image;

quantizing the floating point DCT frequency coefficients;

compressing the quantized floating point DCT frequency coefficients, wherein the compression of the quantized floating point DCT frequency coefficients is lossless; and

storing the compressed quantized floating point DCT frequency coefficients.

2. The method of claim 1 , further comprising:

converting the floating point image data values of the HDR image from a first color space to a second color space prior to determining the floating point DCT frequency coefficients.

3. The method of claim 2 , wherein the second color space is a Y′ CbCr space comprising luma values and chroma values.

4. The method of claim 1 , wherein the mapping function comprises a piecewise power function and a piecewise logarithmic function, the piecewise power function mapping a first range of input values using a power function, the piecewise logarithmic function mapping a second range of input values using a logarithmic function.

5. The method of claim 1 , wherein quantizing the floating point DC frequency coefficients comprises:

quantizing a floating point DCT frequency coefficient to a floating point quantized DCT frequency coefficient within a per-frequency coefficient absolute error tolerance of the floating point DCT frequency coefficient.

6. The method of claim 5 , wherein the per-frequency coefficient absolute error tolerance is computed by multiplying a base error tolerance by a per-frequency relative error tolerance.

7. The method of claim 6 , wherein the per-frequency relative error tolerance is based on a DCT quantization table.

8. The method of claim 7 , wherein the per-frequency relative error tolerance is computed by normalizing a DCT quantization table.

9. The method of claim 5 , wherein quantizing the floating point DCT frequency coefficient further comprises:

selecting a floating point value having a minimal number of bits set to one.

10. The method of claim 9 , wherein quantizing the floating point DCT frequency coefficient further comprises:

selecting a floating point value having the least quantization error.

11. The method of claim 5 , wherein quantizing the floating point DCT frequency coefficient further comprises:

selecting a floating point value having a maximum number of bits set to one.

12. The method of claim 11 , wherein quantizing the floating point DCT frequency coefficient further comprises:

selecting a floating point value having the least quantization error.

13. The method of claim 5 , wherein quantizing the floating point DCT frequency coefficient further comprises:

looking up a pre-computed table.

14. The method of claim 1 , further comprising:

decompressing the compressed quantized floating point DCT frequency coefficients, wherein the decompression is lossless;

converting the decompressed coefficients to a set of decompressed floating point image data values of the HDR image using a floating point inverse discrete cosine transform; and

mapping the set of decompressed floating point image data values of the HDR image from the perceptually linear space to the light linear space using a second non-linear mapping function, wherein the second non-linear mapping function is the inverse of the first non-linear mapping function.

15. A computer-enabled method of decompressing a lossy compressed high dynamic range (HDR) image, the method comprising:

obtaining compressed quantized floating point discrete cosine transform (DCT) frequency coefficients for the HDR image;

decompressing the compressed quantized floating point DCT frequency coefficients, wherein the decompression is lossless;

converting the decompressed coefficients to a set of decompressed floating point image data values of the HDR image using a floating point inverse discrete cosine transform; and

mapping the set of decompressed floating point image data values of the HDR image from the perceptually linear space to the light linear space using a second non-linear mapping function, wherein the second non-linear mapping function is the inverse of the first non-linear mapping function.

16. A non-transitory computer-readable storage medium comprising computer-executable instructions for lossy compressing a high dynamic range (HDR) image, the instructions comprising:

mapping floating point image data values of the HDR image from a light linear space to a perceptually linear space using a first non-linear mapping function, wherein a value in the light linear space is proportionate to the power of a light source, and wherein a value in the perceptually linear space is proportionate to a visual perceived brightness;

determining floating point discrete cosine transform (DCT) frequency coefficients of the floating point image data values of the HDR image;

quantizing the floating point DCT frequency coefficients;

compressing the quantized floating point DCT frequency coefficients, wherein the compression of the quantized floating point DCT frequency coefficients is lossless; and

storing the compressed quantized floating point DCT frequency coefficients.

17. The computer-readable storage medium of claim 16 , further comprising instructions for:

converting the floating point image data values of the HDR image from a first color space to a second color space prior to computing the floating point DCT frequency coefficients.

18. The computer-readable storage medium of claim 16 , wherein instructions for quantizing the floating point DCT frequency coefficients comprise instructions for:

quantizing a floating point DCT frequency coefficient to a floating point quantized DCT frequency coefficient within a per-frequency coefficient absolute error tolerance of the floating point DCT frequency coefficient.

19. The computer-readable storage medium of claim 16 , further comprising instructions for:

decompressing the compressed quantized floating point DCT frequency coefficients, wherein the decompression is lossless;

converting the decompressed coefficients to a set of decompressed floating point image data values of the HDR image using a floating point inverse discrete cosine transform; and

mapping the set of decompressed floating point image data values of the HDR image from the perceptually linear space to the light linear space using a second non-linear mapping function, wherein the second non-linear mapping function is the inverse of the first non-linear mapping function.