Based On The Color And Shape Features Of The Image Retrieval Technology Research And System Design And Implementation

This paper utilizes image processing, computer vision and database technology to make a study of the key technology in content-based image retrieval (CBIR). When extracting the images'feature, we major research on how to fully and effectively describe the image color, shape features, as the image itself is complex and contains mass information. The color feature is mainly based on color histogram and color moments. Without affecting the accuracy of the features described, for the purpose of reducing histogram dimension, we improve the existing color histogram, and then integrate it with color moment information. Taking the retrieval effectivity into account, we construct the color similarity coefficient of two images by merging images'average color similarity coefficient with images'main color similarity coefficient to filter the image database by setting an appropriate threshold, which reduces the image library's searched space and improves the retrieval rate. After searching an image set which roughly meets the requirements, we focus on studying the color characteristics of the matching and similarity measure in the filtered images set.In the shape feature, we mainly study the dual-threshold problem of Canny edge detection algorithm and color based-on multi-structure elements morphological edge detection algorithm. After that, we combine five promoted moments with seven Hu invariant moments to constitute shape retrieval vector for images'shape retrieval. The retrieval performance is improved as the addition of some shape details factors.Single feature-based image retrieval often catches one and loses another, so it can't integrate the advantages of each feature. After studying based-on color feature or shape feature image retrieval, we combine the both information, and realize multi-feature integrated image retrieval which filters the image database by merging image's average color similarity coefficient with image's main color similarity coefficient. Here, the shape feature descriptor is image edge orientation histogram descriptor which extracted from color edge detected by based-on multi-structure elements morphological edge detection algorithm.Through studying and analysising image retrieval system model, we implement a prototype image retrieval system with Visual C++6.0. Using this system and a test image set composed by 1000 images selected randomly from corol gallery to validate the above methods. The experimental results show that:these methods can improve the retrieval rate and accuracy of image matching.