Compression-based semantic-sensitive image segmentation: PRDC-SSIS
Masahiro Nakajima, Toshinori Watanabe, Hisashi Koga · 2012
This paper proposes PRDC-SSIS, a new compressibility-feature based semantic-sensitive image segmentation method using PRDC. One of the drawbacks of traditional signal (pixel-color) based image segmentation is the poor capability to capture the semantical information contained in the images. Because the semantic information tends to be carried by a set of neighboring pixels, rather than an individual pixel, we divide the image into patches and classify the patches based on their semantical contents. The crucial problem is classifying the patches into groups of similar patches according to their contents, and so we exploit the compressibility feature vector space of PRDC to accomplish this. An application of this method to an EO-image confirmed the proposed scheme can be carried out without any of the human-tailored target object models required by almost all traditional methods.