A Robust Feature Extraction Arithmetic To Geometric Distortions For Texture Image

With the advances in information technology there is explosive growth of the multimedia databases and the internet, which demands effective and efficient tools that allow users to search and browse through such a large collection. As the databases grow larger, the traditional keyword based method to retrieve a particular image becomes tedious and inadequate. To solve these problems, content-based image retrieval (CBIR) approach has emerged as promising alternative. In CBIR, images are indexed by their own visual contents, so feature extraction and matching become very important components in CBIR. Queries can be based on many features such as texture, color, shape and their combinations. Importance of texture feature is due to its presence in many real world images: for example clouds, trees, bricks, hair, fabrics etc., all of them have textural characteristics.A lot of work has been done on texture analysis, and the majority of existing work assumes that all images are acquired from the same viewpoint. This assumption is not realistic in practical applications. The performance of these methods becomes worse when this underlying assumption is no longer valid. Several authors have proposed rotation invariant texture features. All these rotation invariant texture analysis methods were designed for the classification problems, where the classes are defined a priori. Therefore these methods are not suitable for the retrieval applications, where each database image forms a separate class and must be trained individually. This thesis proposes a robust feature extraction technique for texture image retrieval, the method is robust to geometric distortions. The geometric distortions include rotation, scaling and translation modifications of textures. Compare with traditional methods, our method shows higher accuracy for geometric distortions image retrieval.This thesis is divided into three parts. The first part introduces the image retrieval and the leading feature extraction technique for retrieval, the second part explains the detail arithmetic of our thesis, the third part gives the simulation result and compares the results with the traditional methods. At last we discuss the subsequent work. The creation of our method is that, In the feature extraction process, the log-polar transformed autocorrelation image which expresses regularity and continuity of texture was introduced to eliminatethe entire geometric distortions. In the retrieval process, a set of wavelet packet statistics of the log-polar transformed autocorrelation image was used to measure similarity of texture images which is different with the traditional methods that uses a set of wavelet packet statistics of image itself. In the simulations, the robustness to geometric distortions of our method was verified by L2 distant and higher accuracy compared with other method for retrieval was demonstrated in the experiments.