Design And Implementation Of Image Retrieval System Based On Markov Steady State

With the development of computer, multimedia and Internet technology, especially the widely used search engines, people can access to larger and larger number of image data. The topic that how to retrieve the required image from the large-scale image database quickly and efficiently became more and more important and challenging. Content-based image retrieval is just the smart and efficient solution to this problem.Content-based image retrieval is the task of classifying images according to the object category it contains. Image Feature extraction and image retrieval are the most critical steps which play an important role to the performance of retrieval. They are also hot subjects of many recent papers. Although many recent researches, for example histograms of oriented gradients (HOG), pyramid histograms of oriented gradients(PHOG) or support vector machine (SVM), have achieved some success, it is still a challenging task and far from satisfactory owning to two parts. One is the insufficient image descriptor and the other is that semantically related images may reside in an embedded manifold and not a linear hyperplane in the feature space.The research starts from the two facets of image feature extraction and fast image retrieval algorithm. In image feature extraction stage, we propose the gradient histogram Markov stationary features to represent the input image which is capable of characterizing the spatial co-occurrence of gradient histogram patterns. In image retrieval stage, the image training and retrieval process is treated as searching for an ordered optimal cycle in the image database by minimizing the geometric manifold entropy of images. Experimental results demonstrate that the proposed framework for image retrieval is feasible and gradient histogram Markov stationary feature apparently outperforms the original HOG descriptor in feature representation.The following highlights the major contributions of the paper:·In image retrieval stage, we extend image gradient histogram based features. We characterize the spatial co-occurrence of gradient histogram patterns by Markov chain models, and finally yields a compact feature representation Gradient Histogram Markov Stationary Feature (GHMSF) through Markov stationary analysis.the GHMSF goes one step beyond gradient histograms since it now involves spatial structure information of both within gradient histogram bins and between gradient histogram bins.·It is difficult to find out the mapping from low-level feature space to high-level semantic manifold and the dimension of semantic space is hard to decide in advance. The retrieval is treated as searching for an ordered cycle in an image database. Thus, we avoid the mapping from low-level feature space to high-level semantic manifolds.·Tabu search is a common solution to the optimization problems. However, picking up the best candidate in this method is very time consuming. In this study, we improve it by using active tabu search to raise the efficiency of image retrieval.