FRACTAL IMAGE COMPRESSION

Abstract

In context of medical imaging or satellite images, the change in conditions is to be observed at different times. There is a fact that the images captured of the same scene or the same object at different times may not have the same orientation. For the sake of automating the observation of the progress or the change in the condition of the object, there needs to be a technique to align the input image to the reference image. The technique used for the above process is known as image registration. The work under this project is divided in two phases. The first phase is about the image registration of the medical images. It is implemented using MATLAB. Second phase of the project deals with the compression of the images. The method of image compression chosen is Fractal image compression. Since the data in medical images are very crucial, It is not appreciable to loose data in compression-decompression process. As in fractal image compression, the decoding process involves the iteration on the image. If we increase the number of iterations, more accurate picture we can get back. The above mentioned fact is the reason to choose fractal image compression for the purpose of image compression. We realize the importance of compression methods in our daily life when we store files in a limited storage space or when we have to send a file on a slower network. In this thesis, the methods of compression of images is dealt with. Again, there are different approaches for image compression among which jpeg is a well known one. We judge the ability of a compression method by the compression ratio it provides. It is noted that for the images having fractal properties in terms of self similarity or the images having similar regions, the image compression method known as Fractal Image Compression can give better compression ratio. The fractal image compression algorithms have a common approach which involves the partitioning of the image into smaller non overlapping square subsections range blocks of predefined size. Then, a search codebook (domain pool ) is created from the image taking all the square blocks (domain blocks) of size double of the range blocks and ultimately for each each range block, the most appropriate domain block is selected from the domain pool. It is noted that that what transformations are required to be performed on the range block to match with the domain block.