Performance of Wavelet Based Compression in TI TMS320C6201 DSP

Image/signal compression with high fidelity reconstruction at very low bit rates (compression ratios around 100:1) is a necessity in many applications in multimedia, digital libraries, and telemedicine. The advantage of using transform coding is that the signals are expressed in terms of orthogonal basis functions and thus have decorrelated components. Wavelet based compression algorithms are used as wavelets have localization in both time and frequency domains. One wavelet based compression algorithm has been implemented in the TI TMS320C6201 fixed point DSP with a 50% increase in speed at the Computer Vision and Image Analysis Laboratory (CVIAL) at Texas Tech University. Wavelet compression algorithms can be improved by designing image specific filters for an application. The algorithm described here creates wavelet filters adaptively at a time when the image becomes available to it. The best wavelet decomposition for a given image is found, creating the “most convenient space” for the follow up algorithm (e.g. EZW [3,9] or AFLC-VQ [7,8] developed at CVIAL). This results in a better overall algorithm’s performance. One of the optimization criteria used during the algorithm’s trial was a maximization of the scaling function energy (the lowresolution domain). Daub 4[4] filters were used for comparison. The algorithm found a 4-coefficient wavelet filter that concentrated 3 times more energy in the low-resolution domain than the Daub 4 filter. This resulted in a lower distortion rate when the EZW algorithm was used to compress the image.