Key Technology Research And Implementation Of Plant Image Retrieval System

Plant is the focus of human’s attention as an indispensable element of human life. With the rapid development of computer vision and machine learning, people increasingly want to be able to use computers to classify、identify the plant images. However, due to the great number of plant species and the variety diversity between the plants makes this task challenging. This paper amis to do some research on this issue.This paper study the image preprocessing, feature extraction, feature matching to identify the leaf and the flower images. Then we propose two solutions for the leaf and the flower images. Also we do the corresponding experiments to test their effectiveness.For the leaf images,,we First use Otsu threshold method to segment the leaves. Then we extract the image features including multi-scale triangular, direction fragment histogram and salient points based features. Finally, with local sensitive hash and Leave out algorithm we complete the fushion and matching of the multi-features.Meanwhile, for the flower image, we first extract several SIFT-based color descriptors, including HSV-SIFT, Hue-SIFT, Opponent-SIFT. On this basis, this paper introduces a polynomial embedded algorithm to enhance the descriptors and to obtain potential characteristics. The final stage of feature fusion and matching we apply an algorithm based on Fisher vector and linear regression.Also, we integrate open source Web framework and implement a system of content-based image retrieval of plants, mainly including plant image feature extraction, feature indexing,multi-image and multi-organ image retrieval and user feedback.