Research On Image Retrieval Technology Based On ORB Features

With the development of information science and technology,digital image data shows a trend of blowout development.How to query the pictures we want from these massive data has become a hot topic in current research.Especially in the field of mobile intelligent devices,due to the limitations of computing power and storage performance,running image retrieval algorithms on these devices will face many difficulties,especially deep learning with reliable performance is difficult to apply to the field of mobile devices easily.In order to solve these problems,this paper will explores an image retrieval method with small storage space and fast calculation speed and high accuracy based on traditional algorithms,which is applied to the field of mobile intelligent devices.The critical part of the image retrieval process is feature extraction and feature coding,but some current mainstream algorithms still have some problems.Therefore,this article starts from these two aspects and combines the direction of the mobile intelligent device we determined to conduct research.The main research works are as follows:1.Since the binary descriptor ORB algorithm is still not effective in the real-time field,an improved ROI-FAST algorithm for filtering the search domain is proposed based on the FAST algorithm.Since most pictures have areas where the grayscale does not change significantly,these areas usually have very few feature points extracted.In order to reduce the search area of the FAST detection operator,this article removes these areas from the FAST search range according to the idea of classification,leaving only the region of interest for feature point extraction.Experiments show that the improved ORB feature extraction algorithm effectively improves the efficiency of the entire matching system without affecting the accuracy of image matching.2.Aiming at the problem of the jitter of the number of ORB features when the contrast of the picture environment changes,this paper proposes an improved FAST algorithm with an adjustable radius based on the study of the FAST algorithm.We found that when we use different size detection templates,the number of detected feature points is also different.Therefore,this paper combines the detection size and the environmental contrast of the image into a linear relationship,so that it changes with the environment contrast to respond to the complex external environment.Finally,the experiment proves the effectiveness of the improved algorithm in this paper.3.Aiming at the problem that the color information of pictures is discarded in the image retrieval process,this paper proposes a new VLAD coding method based on combining color information based on the traditional VLAD algorithm.First,convert the image to RGB and grayscale images,and obtain the image matrix on the three color channels of Opponent's opposing color space according to the formula,and then add the binary color coding based on the three-channel color matrix on the feature encoding stage.Finally,the improved ORB algorithm and the new VLAD coding algorithm are combined to generate a complete image retrieval method.Experiments show that this method effectively improves the utilization rate of color information of pictures and can meet our requirements for image retrieval.The paper contains 43 pictures,8 tables,and 80 references.