Spatial Information Of Content Based Image Retrieval

Image can cover most of the information around us. There are thousands of image information will be stored for waiting to be processed. In recent years, many researchers have focused on the application of image processing. How to accurately describe image content in detail is the focus of researchers work. After decades of research and development of content-based image retrieval technology is the most mainstream method at the present stage, it is a kind of mathematical model through a variety of extracting image features such as color, shape, texture for image recognition and image matching retrieval technology. General matching methods are gray histogram, scale invariant feature transform, bag of features and so onThis article bases on content-based image retrieval technology for further research. We got the improvement from two aspects in view of the two different kinds of image retrieval method.1. For part of the medical images, we put forward the improved method of image retrieval based on Gabor-Zernike. Gabor filter can extract the shape of the image information and is much efficient than other methods. Zernike matrix can get good deal with image texture information. Based on the content of rotation invariant Gabor-Zernike method is combining Gabor and Zernike matrix using the same scale and different point of view to extract the image rotation invariant texture feature. The experimental results show that the relative of the existing image retrieval method has higher efficiency, overcomes the existing technology of precision caused by inaccurate description and falling shortcomings. Also, in noisy case, the image retrieval precision is higher. Relative to the existing image retrieval method has higher efficiency, and the running time is shorter.2. According to some natural images, we put forward the improved method based on bag of features method in image retrieval area. Bag of features image retrieval is one of the most mainstream methods for natural images, but it only considers the local characteristic information and ignores the image information of whole space position which is important of the image. The Hilbert curve is a kind of is able to fill the whole 2D space, and keep the information between points and neighbor a curve. We applied the improved Hilbert curve to strengthen the word bag method of spatial information, and then generate a Hilbert curve tree structure for image matching and retrieval technology. The experimental results show that the improved Hilbert curve tree structure in the scan images have better results and choose path and the average accuracy of this algorithm with a high increase. If we can select the optimized path, theory will also reduce the tree structure in the generated when the layer. The less the layer number is, the more concentrated the information of each layer is. Also, the complexity of the dictionary will be lower, so the algorithm will be more efficient in principle.For different image datasets, this thesis has taken a different improvement method. It is a big problem in the field of image retrieval: according to different characteristics and different image database, there isn't a unified method to evaluate. So the further target is finding a more effective method in all image retrieval area.This paper was supported by the National Natural Science Foundation(No.61373111),the Provincial Natural Science Foundation of Shaanxi of China(No. 2014JM8321) and the Fundamental Research Funds for the Central Universities(Nos. K50511020014,K5051302084).