The Study Of Content-based Flower Image Retrieval Techniques

With the rapid growth of digital image information resources, how to find the image information rapidly and accurately to meet the needs of users from large image database has become a problem to be solved. Traditional text-based image retrieval techniques cannot meet the needs of users, then content-based image retrieval techniques obtained the rapid development in recent years, mainly using low-level features of an image (such as color, texture and shape) for similarity measure.Visual attention mechanism is a significant characteristic of human visual mechanism, some researchers adopt the method of mathematical modeling on computer to simulate this visual mechanism automatically.Then the visual attention model is established. The research of visual attention related to biology, psychology, computer vision, image processing, and other fields. It can be applied to object detection, image retrieval, video image compression, etc.This paper introduces the key techniques and development trend of content-based image retrieval (CBIR). And based on this, the problem of flower image retrieval is studied. The main work is as follows:(1) In studies of the general image retrieval method based on color, introducing the common way to quantify some color space, two kinds of color features (color histogram and color moment), image information entropy and the thought of blocked image. According to this, a kind of image retrieval method based on color entropy and block color moment is put forward, and applied into flower image retrieval, which produces a good result.(2) Discuss the classic visual attention model at home and abroad, combing with the characteristics of flower images, propose an improved visual attention model, the model processes images in CIE L*a*b*color space, replaces the Fourier transform in spectral residual model with binary wavelet transform, approximately estimates the redundant information of images using the average of logarithmic module, and increases the feature of texture roughness. Finally, the saliency maps obtain satisfactory.(3) With study on the saliency maps of flower images, this paper extracts the low-level features based on these saliency maps. In order to highlight the flower region and without losing the image background information, weighted color histogram and weighted LBP histogram based on saliency maps are adopted. On the other hand, in order to get the features of flowers themselves, the salient edges are obtained with saliency maps, and then the gradient magnitude histogram and gradient direction histogram of the edges are adopted. The experimental results show that these features are beneficial to the flower images retrieval. (4) Apply SVM classifier to flower images retrieval, using grid searching technique and five folds cross validation to obtain the optimal parameters of SVM kernel function. Establish the flower image retrieval system. And the performance of different visual attention model and all kinds of low-level visual features are tested and analyzed. Experimental results show that the flower image retrieval algorithm in this paper is ideal and of universal applicability and transplantability.