Image Retrieval Based On Invariant Features

As a kind of media which can represent much information fairly, images are widely used in various fields. How to retrieve relevant images quickly on demand in the enormous digital image databases has become the bottleneck of information processing techniques. So image retrieval has become a research hotspot in the world. Since many image collections are purely indexed or annotated and lead to bad retrieval results, so content-based image retrieval technique is proposed and developed, it has become one of the important research fields of image engineering now.Invariant content-based image retrieval has widely application foreground. Using the rotation, scaling and translation invariant features can get better retrieval results than normal color, texture and shape features. Thus, in this paper invariant features for image retrieval are investigated. We have research of many invariant features and two new methods are proposed to extract RST (rotation scaling and translation) invariant features by combining the integral invariant and the scale invariant keypoints extractor technique. The content of this thesis is as follows:Firstly, this thesis systematically summarizes the research status of invariant features for content-based image retrieval. At the same time, some classical algorithms are also introduced. The purpose is to understand the existing algorithms and provide background of the retrieval methods proposed in this thesis.One retrieval method using RST invariant features based on LPM (log-polar transformation) transformation is proposed. Considering the characteristic of Hans Burkhardt's rotation and translation invariant features and combining with image LPM transformation method and histogram descriptor, a RST invariant feature extraction method is proposed in this thesis. The main idea of the method is: first, some of the image pixel points are randomly selected. According to a certain radius, a local invariant feature descriptor of the points is constructed and the rotation and translation invariant features of an image are calculated. Then we perform log-polar transformation on the original image to convert the scaling and...