A NOVEL HYBRID DWT-QUAD TREE BASED COMPRESSION TECHNIQUE APPLIED TO 2D IMAGES

Abstract

In the recent years, the development in the field of multimedia has drastically improved the quality of image/videos. Now a days very high resolution image/ videos are available. This has brought a boom in the applications such as photography, television, remote sensing, robotics and medical diagnosis. But the high quality of image/videos has contributed to insufficient bandwidth and memory storage. Therefore, it becomes more significant to compress the data to reduce the data redundancy so that the hardware space and transmission bandwidth could be saved. There are two approaches for image compression named as lossless and lossy. Removing Inter pixel and Coding redundancies from image contributes the lossless compression while the reduction in psycho-visual redundancy contributes the lossy compression. Lossless compression doesn’t give any data loss while there is a certain amount of data loss in Lossy compression. In this report, the analysis of the performance of lossless compression algorithms as Run length Coding (RLE), Huffman Coding, Arithmetic Algorithm and Lempel-Ziv-Welch (LZW) Algorithm has been carried out on the basis of the parameters like Compression Factor, Compression Ratio, Saving Percentage, and Compression/de-Compression time. Random images of different probability of zeros have been taken for this purpose. The lossless compression algorithms are applied to the above images and the results are plotted as well as presented in tabular form. Further, the performance of the Lossy compression methods Discrete Wavelet Transform (DWT), Discrete Cosine Transform (DCT) and Quad-tree decomposition method have been analysed with respect to Peak Signal-to-Noise Ratio (PSNR) and Mean Square Error (MSE). A lossy compression algorithm has also been proposed. This algorithm is based on wavelet compressed Quad tree decomposition and parametric line fitting. The image is decomposed into the new images LL, LH, HL, and HH according to the high-pass or low-pass filter applied to the original image in the vertical and horizontal directions. LL, which contains the most of the original image details, is further decomposed in homogenous and nonhomogenous blocks with 4x4 as minimum block size. The fitting process has been applied v to each segment and the fitted data is found out. The process is repeated by taking the new set of breakpoints. Finally the Entropy encoding is carried out. Parametric line interpolation is performed in order to retrieve the image. Images with different colour contrasts are considered so as to test the compression algorithms in more rigorous way. The compression results with respect to each image have been demonstrated qualitatively in tabular form. The MSE and PSNR values are also obtained for each image and compared with the help of Bar charts. The comparative analysis shows that the proposed algorithm gives the improved results as compared to DCT, DWT and Quad-tree decomposition method. All the compression algorithms have been implemented in MATLAB R2009a.