Research On Building Image Recognition Method For Mobile Application

With the rapid development of information, a variety of mobile devices are continuing without end. The development of mobile hardware makes researches oriented mobile terminal become a new hot spots. The research on building image recognition system oriented mobile application is an important technology. Before the rapid development of intelligent mobile phone, the research on image recognition is mainly used in workstation or desktop computer. This way of recognizing image is mainly used in scientific research laboratory, it can take little convenience for people’s daily life. This paper mainly studies the recognition system of building image oriented mobile application. It has important theoretical significance and practical application value for mobile development. In this paper, the main research results are as follows:(1) The traditional image scaling technology scales image without according to the importance of pixel. In order to solve this problem, this paper mainly focus on an image scaling technology which can preserve SIFT features of image. Firstly, gray the image; Then, get the gradient energy and the visual saliency energy of the image, and add them into a new energy as the complex energy function; Get the cumulative energy matrix and minimal energy trajectory matrix with the complex energy function, and get the minimal energy seam; Finally, insert or delete the minimal energy seam to complete the image scaling. Experiments show that, it is feasible and effective.(2)SIFT features is a high dimensional vector, so image feature extraction with traditional SIFT algorithm will take a long time and cost high memory, in order to solve this problem, studied and proposed an improved image feature extraction technique based on SIFT. Firstly, build the scale space of difference of gauss; calculate and get the local extremum points of the image, also known as the key points of the image; Then pinpoint the position and scale of local extremum points, use the stable direction of local image as the direction of the key points with the image gradient histogram; Finally, statistics9seed points with8directions as the feature descriptor, and quantify the feature descriptor to get the final feature descriptor. Experiments show that, it is feasible and effective.(3)It has a low accuracy when matching two image descriptors with traditional Euclidean distance of SIFT algorithm. In order to solve this problem, studied and proposed an improved image feature matching techniques based on SIFT. Firstly, respectively use Manhattan distance, Chebyshev distance, Mahalanobis distance as the distance measurement to complete SIFT feature matching, make comparisons among these distance measurement with a large number of tests, find that Mahalanobis distance is the most efficient distance measurement for SIFT feature matching algorithm, so make Mahalanobis distance as the new distance measurement during feature matching; And then use the K-D tree as the data index to speed up the feature matching; Finally, use the RANSAC algorithm to eliminate the error matching points. Experiments show that, it is feasible and effective.(4)Design and implement the building image recognition system oriented mobile application. In order to improve accuracy and efficiency of the building image recognition system, in this paper, it designs and implements four important modules, such as image feature extraction, image feature database building, image feature matching and image data transmission. Finally, completed the building image recognition system for mobile application. The effect of experiments shows that the building image recognition system is running well.