(A) selective coding algorithm for still images based on an extension-interpolation method and target recognition

Abstract

Various coding schemes such as JPEG, H.261, MPEG have been proposed, standardized, and widely used in many fields. But in most systems using such schemes, the relative importance between the region of interest(target region) and the background region is not considered. So, when the capacity of transmission channel is highly limited or when high compression ratio is demanded, it is difficult to use them because many bits must also be allocated for the background region which is less important than the target region. In this paper, we propose a new it selective coding scheme for still images and efficient coding of arbitrarily shaped image segments for this scheme. First, we segment an input image using morphological filters and a watershed algorithm. Next, we extract the target region from the segmented image using moment Fourier descriptors(MFD) with the database which is assumed to be constructed in advance by a priori knowledge of the target we want. Then, we code the shape information and the texture information of the target region in higher quality than the background region. Simulation results show that the proposed coding scheme provides better performance than the conventional JPEG standard in the same compression ratio.