J. Froment, "Image compression through level lines and wavelet packets," in Wavelets in Signal and Image Analysis, A. A. Petrosian and F. G. Meyer, Eds. Kluwer Academic, 2001.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....coding the sketch, we wish to code the residual texture t = f Gamma c using standard wavelet techniques. Unfortunately, residual artifacts created by wedgelet approximations pose many of the same difficulties for wavelets that edges present. A refinement can be made to the residual image [4], masking out any possible artifacts located near edges in the cartoon sketch. This makes the resulting residual image much easier to code using wavelets, and we see a noticeable improvement in visual quality over standard coding techniques. The destruction of information, however, prevents the ....

J. Froment, "Image compression through level lines and wavelet packets," in Wavelets in Signal and Image Analysis, A. A. Petrosian and F. G. Meyer, Eds. Kluwer Academic, 2001.

....implementation of our coder with a representative zerotree (EZW) wavelet coder [2] Most importantly, although our techniques tend to degrade PSNR performance, we note a signi cant increase in visual performance, with sharply de ned edges and minimal ringing artifacts. In related work, Froment [7] and Dragotti [8] have developed techniques for explicitly extracting contours from images, while Mallat and Pennec have proposed bandelets [9] as a contour estimate that explicitly deals with nearby artifacts. Meyer [10] has further justi ed that this sort of mixture modeling of image features ....

....1. Assuming that we wish the cartoon to provide the only coded information about the edges, then we should discard these di erence artifacts before coding the residual. A simple technique is to mask the edge artifacts out of the texture image, setting t = 0 near all cartoon edges (as in [7]) With much of the high frequency information removed from the image, a standard wavelet coder eciently codes the residual. In this paper, to eliminate the ridge artifacts in the residual image, we implement a tapered masking scheme. First, for each pixel (x; y) in the image, we compute the ....

[Article contains additional citation context not shown here]

J. Froment, \Image compression through level lines and wavelet packets," in Wavelets in Signal and Image Analysis, A. A. Petrosian and F. G. Meyer, Eds. Kluwer Academic, 2001.

....that edges present. These artifacts typically resemble tall, thin ridges which have a geometric structure similar to the edges themselves. Coherency must be preserved among their quantized wavelet coefficients in order to prevent ringing. It is possible to make a refinement to the residual image [8], masking out any possible artifacts located near edges in the cartoon sketch. This makes the resulting residual image much easier to compress using wavelets, and we see a noticeable improvement in visual quality over standard coding techniques. The destruction of information (during masking) ....

J. Froment, "Image compression through level lines and wavelet packets," in Wavelets in Signal and Image Analysis, A. A. Petrosian and F. G. Meyer, Eds. Kluwer Academic, 2001.

....fewer bits were available to code textures, though, the ultimate PSNR was lower than the standard zerotree technique. As demonstrated by the above experiment, for a natural image coder which uses geometrical coding, the residual information near coded edges presents particular challenges. In [16], Froment demonstrates the effectiveness of coding this type of residual using wavelet packets. We seek, however, an approach which is more closely tailored to the geometry of the contour. In the next section, we present a simple technique which avoids the challenge of compressing these ridge ....

....of the locations of the edge approximations, and hence of the possible ridge artifacts. A simple method for avoiding the difficulty of wavelet based compression of these artifacts is to mask the possible artifacts out of the texture image, setting e t = 0 near all coded cartoon edges (as in [16]) This removal of information (which may not be recovered by the decoder) has a significant impact on the PSNR. It also discards legitimate texture information close to edges, which cannot easily be separated from edge artifacts. The resulting residual image, however, may be compressed much more ....

J. Froment, "Image compression through level lines and wavelet packets," in Wavelets in Signal and Image Analysis, A. A. Petrosian and F. G. Meyer, Eds. Kluwer Academic, 2001.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC