Discrete Cosine Transform (DCT) is an effective image compression method based on the orthogonal transform. However, it is well known that the visual quality of encoded images using DCT is deteriorated around edge at the lower bit rate. On the other hand, Independent Component Analysis (ICA) can derive an inherent set of bases which are correspond to local features for a given input image. When ICA has been applied to image compression instead of DCT, its coding performances is inferior to the conventional DCT method. This paper presents ICA-DCT hybrid typed image coding. In our method, each signal of small region in the input image is transformed by ICA or DCT appropriately in the viewpoint of the reduction of the bit rate. In order to solve the problem that a set of ICA bases in the proposed method has the higher redundancy, the quantitative measure to extract the important ICA basis is defined as the fluctuation of bit rate by the replacement of ICA basis using a set of DCT bases. It is seen in our experimental results that the coding performance of the proposed method is improved in about 0.5 [dB] compared with the conventional DCT method for real world images.
View full abstract