Based On Evolutionary Algorithms Curl DCT Image Compression Research

The amout of data in image is so big that the image must be compressd before storaged andcommunicated. So the image compression is an unavoidable step before storage andcommunication. And it is an important part of image communication technology. Transformcoding is a core part of the image compression. Currently, JPEG is the most commonly useed inimage compression coding standard, which adopts the discrete cosine transform (DCT). However,the compression coding based on DCT will lose much high frequency information, so that it willmake the image blurry. Warped DCT combines the warped principle and DCT. Through anall-pass filter, input signal is warped so that the warped signal is more suitable for DCT coding.With this warping, it can overcome the shortcoming of image blurred.The major work and innovations in this dissertation are as follows:(1) A warped DCT image compression scheme based on evolutionary algorithm is proposed.In this scheme, the warped DCT image compression method is combined with evolutionaryalgorithm. By the global optimization of evolutionary algorithm, the optimal frequency warpingparameter is selected adaptively to overcome the shortcomings of the traditional warped DCTmethod. Through the optimal frequency warping parameter, this method can improve theperformance of warped DCT image compression method effectively. With the common standardimages compressed, the fist experiment shows that the proposed method can improve the peaksignal to noise ratio effectively. This experiment verifies the effectiveness of the proposed method.With one hundred images in open image database of University of Washington compressed, thesecond experiment shows that the proposed method can improve the peak signal to noise ratio,and signal to noise ratio while reduce the root-mean-square error effectively. This experimentverifies the versatility of the proposed scheme.(2) According to the image frequency distribution, a heuristic method to initialize thepopulation is proposed. Because the image energy focuses on the low frequency part, a range ofparameter is selected to initialize the population, which can improve the convergence speed andensure the diversity of the population.(3) According to the characteristics of the frequency warping parameter, real coding is usedto encode the solution of problem and a kind of optimized mutation operator is designed.Through this method, the cumbersome coding conversion steps of traditional evolutionaryalgorithm are removed, the computation and storage overhead are reduced, the encoding andevolution process are simplified, and the convergence speed of algorithm is accelerated.