Region-based Image Retrieval Using Multi-feature Integration

Content-Based Image Retrieval can be divided into two parts,one is Global Features-Based Image Retrieval,and the other is Region-Based Image Retrieval. The former is simpler than the latter, but unfortunately it will produces ambiguity,which means it will make two completely different images have similar features. Region -Based Retrieval applies image segmentation to decomposing an image into regions,and uses region visual feature to represent and index the image.It can reduce the gap between low-feature and high-level semantic concepts and it is closer to the human perception.The thesis mainly studies the related technologies of Region-Based Image Retrieval,and the main research results of this dissertation are summaried as follows:First,in the thesis a Wavelet Multi-scale Analysis based image segmentation algorithm is proposed,in which we applies the Wavelet Transformation to the retrival of a series approximations of the original image. Then the initial segmented regions were achieved by segmenting the largest scale image,and we could got the segmented results of low level scale image of segmented largest scale image. By repeating this process, the image nested segmentations at different levels could be constructed.So,the new method reflected the multi-scale characteristics of real world and reduced the abuse of segmentation.Second,in the thesis the integrated region matching(IRM) is impoved by usage of valid region,we don't calculate the significant coefficient if it's invalid region,which reduce the computational complexity and retrieve time.Third,in the thesis,a image retrieval method is proposed which is based on Multi-scale segmentation and improved IRM.Experiment results indicate that our method can retrieve image efficiently.Finally,the work is summed up and the further research direction is pointed out.