Based On Gabor Transform Visual Feature Extraction

Feature extraction technique has been concerned about the researchers,because their applications are broad,too many method can be used, moreovereach Feature extraction has its own merits, therefore it is very difficultto choose one universal feature extraction method. With the developmentof people living material and production, feature extraction of therequirements are increasingly high, the researchers hope to designfeature extraction algorithm, and this algorithm can be used as far aspossible in quite widespread domain. Introducing signal analysis,theimage feature extraction technology has more methods to be applied, Gabortransformation as a tool for signal analysis, is widely used in digitalimage feature extraction. The Gabor wavelet is multi-resolution andmulti-channel filter,and cover the frequency range to be quite broad.Gabor wavelet, meeting people's request, is widely used in featureextraction,and the improved methods of Log-Gabor transformation has beenwidely application,too.In the life more and more biological recognition technology, in themilitary more and more nobody detection technology,we need a featureextraction method that adapt to changing light environment,this featureextraction method not only to adapt to changing light conditions, butrequired feature extraction accuracy, operator convenience. Now peopleuse some of the feature extraction methods, such as Roberts, Prewitt,Sobel and Canny, etc. Although the calculation speed of these methods isquick,but the application scope is relatively small, the quality offeature extraction is relatively low, these methods can not meet the demanding of the people. Gabor transformation and Log-Gabortransformation in complex lighting environment, feature extractionresults are also affected. In the research,Oppenheim and Lim discoverythat image has the phase information, this phase information is differentfrom amplitude information that is used in traditional feature extractiontechniques, it is difficult to influence in lighting changingenvironment,therefore feature extraction method base on phase congruencyhas been proposed. In traditional methods of feature extraction base onphase congruency,using Log-Gabor transformation of real and imaginaryparts to complete Hilbert change, and then calculate the value of localenergy and local magnitude,make both values of local information to takefinally complete the algorithm. This process of computation consume largememory, the speed of computation is slow. Feature extraction base on phasecongruency also has very significant advantages,the method is difficultto influence in lighting changing environment, and result of featureextraction meet requirement of people.This article proposed phase congruency algorithm based on the Riesztransformation .This algorithm retained the merit of tradition featureextraction based on phase congruency, moreover compares the traditionalphase congruency method based on Hilbert transformation, the memoryconsumed is reduced. The Riesz transformation and the Hilberttransformation have similar properties, therefore may introduce the phasecongruency in the Riesz transformation space. Compared with traditionalphase congruency which use 2D Log-Gabor transformation as a scale functionin Hilbert transformation space, in Riesz transform space using 1DLog-Gabor transformation as scaling function only have result such aslocal direction, local phase and local amplitude, so we can not use phasecongruency without local energy and sum value of amplitude. In this paper,using different wavelengths of 1D Log-Gabor transformation as scalingfunction in same Riesz transformation space, can establish new space in which compute phase congruency. We infer phase congruency algorithm invector space. By comparative experiments, optimize of algorithm, andachieved good results in the simulation experiment.In addition, this article also completes following work:(1) Study extensive use of feature extraction model ,selects Gaborwavelet algorithm based on the signal analysis.(2) Designs Gabor filter, compared with other feature extractionalgorithm,Gabor wavelet has multi-resolution and multi-channel, and toamplitude of images has a good response ,but in the practical application,discovered that Gabor is not completed band width filter, when thebandwidth of frequency is bigger than the center frequency of 1 / 3, realpart of Gabor transform emerge DC component that influence featureextraction;(3) Apply Log-Gabor filter in experiment ,compare the Log-Gabortransformation and Gabor and verify that Log-Gabor transformation asbandwidth filter is better than Gabor transformation. Log-Gabortransformation is asymmetry in the frequency domain,and it has broaderextension of the frequency domain,enabled its performance to haveenhancement。Not only in theory more conformed to the humanity visionsystem,but frequency domain extensions makes the filter computationreduced. However, step changes in brightness of the image do not meet therequirement of researcher;(4) Use Log-Gabor transformation as a scale function, complete thefeature extraction algorithm based on phase congruency. Through researchdiscovery phase information in image universal existence, and has theamplitude information incomparable superiority , amplitude in imageinformation has effect in light condition ,but the phase information isbeen relatively small influence at light change condition, image that hasbrightness step change,can use the feature extraction base on phasecongruency method. Although the feature extraction get satisfactory result of image features, but the algorithm consume memory and run moreslowly.Above these works are aimed at the realization of a universal featureextraction, which is not influence in light change condition. Phasecongruency base on Riesz transformation can use in different lightcondition, but there are aspects of noise interference, from now on wewant to complete the following work:(1) Analysis noise's origin,estimate and compute value of noiseinfluence;(2) Using feature extraction of PC in Riesz transform space, hope toachieve in some simple applications such as edge detection, textureclassification,and so on;Use result information of phase congruency basedon Riesz transform in image registration.In this paper, phase congruency based on Riesz transformation is notperfect, we need to continually improve the method, the author knowledgeis limited ,and hope that readers and experts give advice and guidance.