Digital image compresion techniques

ABSTRACT In some science branches it is required that images are processed and stored into memory. However, 65536 byte of memory is required for an image of 256x256 pixel size to be stored without any processing. This causes serious memory problem if we have lots of images to be stored. In order to overcome this problem different image compression techniques are used. In this thesis, the most popular image compression techniques of nowadays were investigated. It is seen that Run-Length Coding Algorithm is successful at low compression ratios. If the compression ratio is increased the algorithm becomes unsuccessful, and horizontal lines, which are undesired covers the image. The Discrete Fourier Transform and the Discrete Cosine Transform are better than Run-Length algorithm at many different compression rates. Nowadays, the other mostly used compression technique is the Vector Quantization. In this technique, the image is divided into blocks of small images. AH blocks are entered into a training algorithm. Using this training algorithm, a codebook which can represent the image is obtained. The compression is achieved by using this codebook. By using this codebook at the reconstruction process, the image becomes very close to the original image. The Hierarchical Finite State Vector Quantization (HFSVQ) is improved version of Vector Quantization technique. In this technique, the image is divided into blocks of different sizes. Training algorithm is applied to the each block of group. By using the codebook the compression process is achieved. It is noticed that this algorithm has perfect results when the original image has large areas of constant gray level. In this thesis HFSVQ algorithm was applied on different biomedical images (MRI and BT images). In this algorithm the sizes of the blocks change. Large blocks were used toİİİ represent the low contrast area of the image so high compression ratios were obtained. Small blocks were used to represent the high contrast area of the image so low compression ratios were obtained. Since the backgrounds of all the biomedical images have large areas of constant gray level, it is seen that HFSVQ algorithm is very suitable for biomedical images. Finally, HFSVQ algorithm was applied on different MRI and BT images and very low mean square errors were obtained at high compression ratios. It is seen that HFSVQ algorithm which is used to compress the biomedical images is better than mostly other techniques which have been used so far.