The Research Of Karhunen-Loeve Transform And Its Application For Image Compression

Karhunen-Loeve transform is the best transform in the sense of minimum mean square error, can completely disarm relativity between image pixels. But its transformation matrix varies with the image content, is difficult to solve and its data is very large. K-L transform is used in addition to reduce the spectral correlation of multispectral remote sensing image, most is limited to theory research in the field of image compression currently.Present the theory of K-L transform in detail, research and analysis using several commonly used methods of neural networks to achieve K-L transform. Then on the basis of studying of K-L transform deeply, introduce methods to improve it in allusion to its transformation matrix's shortcomings. Then the transformation matrix is easily to solve, and its data is very small than the whole image's. In this way, K-L transform can be applied to the digital image compression. Using optimized K-L transform as the core, with quantization coding and entropy coding, introduce a complete digital image encoding/decoding system based on K-L transform. First gave the overall framework of the system, then divide the system into three functional modules: K-L transform/inverse transform module, quantization encoding/decoding module and entropy encoding/decoding module, Describe the realization process of each module in detail, final test the performance of the system.The system performance testing shows that the image coding system has a good performance in digital image compression. Its compression performance is better than JPEG compression standard which is most commonly used current.