The Research Of Image Processing Based On Local Invariant Features

Improvement of image matching technology makes it more and more integrate into the people’s daily life. The simple definition of Image matching is a technology which two correlation pictures is registering. The feature based on images matching are the most common. Its main task are obtained steady characteristic representative from the multiple images to be measured, described for the proposed feature of selection of parameters by using a variety of mathematical methods, and then built the corresponding similarity description between multiple points. Because of the photosensitive material of the sensor, imaging principle, the measured image angle, and the interference of measurement environment, these are increasing the difficulties of image matching. With the rapid development of the technology of image matching, feature matching is becoming mainstream, it not only can overcome the above shortcomings, but also make the feature points what need to be detected greatly reduced, furthermore, has the ability of anti-jamming and good adaptability.SIFT feature matching and SURF feature matching are two more representative feature matching methods, they have good robustness to noise and image transforming. SIFT feature matching has the advantages of compressed information quantity, high precision, but this method also has some insufficiencies such as large amount of calculations, time-consuming, the object to be measured position strictly. SURF feature matching uses the rectangular filter instead of two order Gauss filter kernel, the matching process is accelerated, as a result of the main direction is more relying on the gradient direction of the local pixels, This will produce more errors. So it is the main target of this papers to find a method that can extract a representative feature vector quickly and stably, form an effective description, complete similarity measurement.The main contents of this paper include:(1) Selecting the hardware system based on the basic theory of Machine vision and completing the hardware system;(2) Using Visual Studio 2005 as the development environment, completing the program design of software system, realizing of statistics of measured object data and analysing of experimental results of measured object data;(3) Using Microsoft Corp specialized for detecting visual sensor of open source function library: Open CV(Open Source Vision Library) to study of image preprocessing, image positioning, feature matching detection algorithm;(4) Using the detection system to test the different object to be measured, verifying the accuracy and feasibility of the system, completing the analysis of experimental results.The experiment proves: the proposed feature matching method in this paper are more satisfied with accurate parameters estimation and requirement of the real-time production among the SIFT feature matching or SURF feature matching; it can be measured to make digital images accurate judgments and recognition in the premise of ignoring the image position. The high efficient matching algorithm of this paper is for the practical follow-up study to lay the foundation.