Iris Feature Extraction And Classification Based On Gabor Filters

This paper adopts the method based on Gabor filters to extract the feature ofiris and classify them.First, the paper describes the overall design and working principle of the irisacquisition instrument. In the pre-processing stage,by using the correlation of irisimage equipment,can we determine a point in the pupil.The correlation is that thereis an image of infrared lightin the pupil. Then, starting from the pupil detectedgradation mutation point using the Sobel operator to the four directions, a total offour points may determine. According to the principleof three non-collinear pointscan determine a circle, a total of four circles can be determined, finally determinethe iris inner circle. Using thesimilar method, we can determine the outer circle ofthe iris.Then by using the polar coordinate transformation and bilinear interpolationmethod, normalized iris image.At last we use the histogram equalization method fornormalized image enhancement. Before histogram equalization, we need toeliminate the impact of uneven illumination, etc. The original iris image is dividedinto several small cubes. Take the minimum gray value of each block as thebackground intensity. Then, each piece of block subtracts the minimum gradationvalue of the piece corresponding to.In the stage of feature extraction and feature vector similarity calculation,thispaper uses a set of self-similar2D-Gabor Filters to filter the iris image. Then we usethe result of filtering to code according to the amplitude and phaserespectively. Forthe phase encoding part uses the Hamming distance to calculate the similaritydistance and normalized,for the average coding part adopts Euclidean distance tocalculate the similarity distance and normalized. Finally, we take the sum of twodistances as feature vector distance. The reason why the result of calculatingforeach distance need normalization, is in order to eliminate their dimension.Thesignificance of this method is that, according to the phase encoding preserves thephase information of original image, according to the magnitude mean coding keeps the energy information of image. The two complement each other, similar degree offeature vector can be more effective.When determining the parameters of Gabor Filters, this paper adopts particleswarm optimization algorithm (PSO) to optimize the Gabor parameters and resultswere very satisfactory. The classification method used in this paper is the k-nearestneighbor algorithm. The method is simple to implement, and better classification.Through the improvement of the k-nearest neighbor algorithm, it does not affect thecorrect identification rate and false reject rate, effectively reduce the error rate.