<title>Progressive perceptually transparent coder for very high quality images</title>

Abstract

In the perceptually transparent coding of images, we use representation and quantization strategies that exploit properties of human perception to obtain an approximate digital image indistinguishable from the original. This image is then encoded in an error free manner. The resulting coders have better performance than error free coding for a comparable quality. F urther, by considering changes to images that do not produce perceptible distortion, we i d e n tify image characteristics onerous for the encoder, but perceptually unimportant. One such characteristic is the typical noise level, often imperceptible, encountered in still images. Thus, we consider adaptive noise removal to improve coder performance, without perceptible degradation of quality. In this paper, several elements contribute to coding e ciency while preserving image quality: adaptive noise removal, additive decomposition of the image with a high activity remainder, coarse quantization of the remainder, progressive representation of the remainder, using bilinear or directional interpolation methods, and e cient encoding of the sparse remainder. The overall coding performance improvement due to noise removal and the use of a progressive code is about 18%, as compared to our previous results for perceptually transparent coders. The compression ratio for a set of nine test images is 3.72 for no perceptible loss of quality.

<title>Progressive analysis-based perceptually transparent coder for still images</title>

“…The rectangular pyramids are slightly worse, and are worse even than the corresponding DPCM results. These results seem to be in con ict with results we h a ve previously presented, 1 but we are presenting di erent DPCM results here. In the prior work, we presented results obtained by Hu man encoding the entire DPCM predicted image.…”

“…The hope is that analyzing the lower resolution images, which in some sense supply non causal information about the signal, will lead to better predictions than standard, non analysis based techniques, such a s D P C M , resulting in smaller errors and a more compressible image representation. In previous work, 1 we used bilinear interpolation and a directional interpolation strategy as the basis for the image pyramid and obtained encouraging results. The results obtained using the directional interpolation were 9% better than those obtained using DPCM.…”

“…Using bicubic spline patches, we then determine, from the fI 1 (j k)g, a smooth surfacê I(x y) t h a t i n terpolates these sampled values and is continuous at the patch boundaries. 1 The approximation, I(m n), of the original image is obtained by subsamplingÎ(x y) a t x = m=S and y = n=S, where S = 8 is the subsampling factor we h a ve c hosen to exploit visual masking of errors in the active areas of the image.…”