| With the rapid development of computer vision,the number of digital images showed explosive growth.As a branch of digital image processing technology,image matching algorithm has been widely used and the demand for its calculation speed is also getting higher and higher.Image matching process is generally divided into three steps: feature point extraction,feature point matching,and elimination of error matching point.This paper mainly completes the improvement of feature point matching algorithm and error matching point culling algorithm,which improves the computing efficiency.The feature point extraction algorithm used in this paper is SURF algorithm,the feature point matching algorithm is KD tree algorithm instead of the traditional linear scanning algorithm,and it has also been improved to reduce the computation complexity by optimizing the query path and considering the data node itself.For the elimination of false matching points,RANSAC algorithm is used,we improved the instability of the original matching matrix and added the threshold judgment to improve the accuracy of matching.The proposed image matching algorithm was applied on the automated testing of mobile phone interface.The existing test frame is used to get the image to be measured,then these image were matched with the standard image,the distortion of text and the distortion of image resolution can be found according to the matching result so as to save the labor costs.The method divides text area and image area according to the region characteristic of the mobile phone interface,and adopts the improved image matching algorithm to match the corresponding area block in the image to be measured and the standard image respectively.If the initial threshold condition is satisfied,the peak signal to noise ratio(SNR)is used to judge whether or not there is a problem of resolution distortion,finally,the interface that does not meet the matching condition is judged as the error interface.The experimental results show that the accuracy of the proposed method can reach more than 95%.Because of the large number of images to be measured,we used the Hadoop as the distributed computing system.Firstly,the whole image matching algorithm is designed and parallelized.Then the HIPI framework is used to design the interface for image data processing and the key-value format of Map / Reduce function.By comparing the time efficiency of single cluster and distributed cluster,we can see that the larger the data volume,the more obvious the advantage of distributed cluster. |
