Analysis of DCTune Yu Kee Lim, Psych 221

What is DCTune Effect of Luminance Masking Effect of Contrast Masking In Conclusion References

JPEG Compression JPEG compression is a standard image compression algorithm that has become well-established. The steps involved are, for each 8x8 block obtain the 8x8 DCT coefficients from the standard DCT matrix quantize this coefficient matrix with a quantization matrix to reduce amount of data to be stored perform lossless encoding on this quantized matrix to obtained a compressed representation of the image The important thing to note is that the quantization matrix is not fixed. This allows different quantization matrices to be used on different images. The standard so far has been to compute the quantization matrix based on experimental results on human visual sensitivity under some factors such as viewing distance and display luminance. Thus the standard compression algorithm doesn't take into account regional variances within an image, rather taking the evaluating the image on a whole. The same quantization matrix is applied to the whole image. What is DCTune DCTune takes this process one step further. As opposed to the standard image independent approach, DCTune performs a compression that is more image dependent. This is an algorithm proposed by Andrew Watson at NASA. Since each image is unique, it would make sense to compute a quantization matrix that is optimal for that particular image. Each image has local variations such that having different quantization matrices that is adapted for individual regions of the image would probably improve the compression. By using different quantization matrices for different regions of the image, there could concievably be better performance because of different human eye sensitivities to vision properties. The two most important local variables that this handles is luminance, and local spatial frequency and orientaion. With regards to luminance, when a region of an image is brighter than the others, the human eye becomes less sensitive to local variation as the contrast has effectively been reduced. Those regions can therefore tolerate greater loss of information, what the author refers to as luminaace masking. The standard jpeg compression only takes into account the average luminance of the entire image, neglecting the possible differences in luminances locally. Another limit of the human eye is its reduced sensitivity to an image component in the presence of other image components of similar frequency and orientation. This is referred to as contrast masking. The DCTune algorithm computes a quantization matrix for each region, taking into account the two factors mentioned above. This allows the image to be quantized more at regions where the human eye will not notice, increasing compression overall. Here, I shall evaluate the performance of DCTune as compared to the standard jpeg compression in terms of these two factors DCTune takes into consideration to determine if it performs as theorized. I used the program written by Andrew Watson which implements DCTune as well as a jpeg compression program obtained from the JPEG Homepage. The links to these two sites are listed in the reference section. Note that the two factors that DCTune really only looks at the intensity levels of images concerned, and does not concern the color information of the image in any way. Thus I only evaluated grayscale images.

                                               

What is DCTune | Effect of Luminance Masking | Effect of Contrast Masking | In Conclusion | References

yuke@leland.stanford.edu