| With the development of Internet plus, a new generation of information technology industry represented by Big Data is developing rapidly. How to search image from a large image database quickly and accurately become a very meaningful and challenging topic. Generally speaking, image retrieval could be mainly divided into two categories: text-based image retrieval(TBIR) and content-based image retrieval(CBIR).Feature extraction is the most important part of CBIR research. Image features consist of low-level visual features and high-level semantic features. Low-level features include global features and local features and what is more, they are the base of high-level features. Comparing with global features, local features are more capable to detect and identify objects in image. Therefore, this paper will focus on the features those are invariant to scale, illumination, rotation, noise change, etc.In order to extract image local invariant features, two parts will be discussed in the paper: key point location research and local invariant feature descriptor research. The first research mainly analyzes the visual invariant between SUSAN corner point and scale space extreme point. Experiments have shown that scale space extreme point detection method is faster and more stability on point numbers than SUSAN corner point detection.Local feature descriptor research is based on extreme points in scale space. Two local invariant features S-SIFT and CC-SIFT based on SIFT will be proposed in the paper. Both of them have a lower dimension for using circular neighborhood instead of square neighborhood in SIFT. The results have shown that improved S-SIFT and CC-SIFT features are as stable as SIFT on scale, illumination, viewpoint, rotation, noise, blur invariant. Also S-SIFT has a best distinguishable ability and CC-SIFT costs least time on feature extraction and matching.In the end, a simple CBIR system using SIFT, S-SIFT, CC-SIFT three invariant features will be shown in the paper. |
PDF Full Text Request