Robust Feature Extraction Arithmetic Based On Parametrical Autocorrelation For Texture Image

With the development of the multimedia and computer network, the digital image capacitance is growing as an explosive speed all over the world. The digital images contain a lot of useful information, but as these images distribute all everywhere, these useful information can not be accessed and used. So it demands a fast and effective technology that can lookup and access the image, this is the so called content-based image retrieval (CBIR). In the CBIR, feature extraction and matching become very important components. And image character includes color, texture, sharp and interspaces. Because texture can show the characters of the image, it has a very large application and research in mode-identify and computer-vision.Lots of work has been done on texture analysis, but these methods are not robust to geometry distortion such as rotation, scale and translation. Several authors have proposed some robust feature extraction technique, using auto-correlation and log-polar transformation. But it still had some problem such as "narrow value region", "exist the decimal fractions", it could not eliminate the influence of shift in log-polar reference frame. This thesis proposes a robust feature extraction arithmetic based on parametrical autocorrelation for texture image, which is robust to geometric distortions. The creation of this method is that: in the feature extraction process, by introduced the parameter, it resolve problems of "narrow value region" and "exist the decimal fractions"; it introduced auto-correlation after log-polar transform to eliminate the influence of shift in log-polar reference frame.