Research Of Iris Recognition Based On Wavelet Transform Combined With Rough Sets Theory

Along with the development of modern informationalized society, the requirement of more accurate and reliable on the identification becomes higher, and identification methods based on biometric identity have been widely used. Compared with face, fingerprint and other biological characteristics, iris identification, which has higher reliability, stability and non-invasive, is considered the most prospects and safe technology and has become a research focus in the biology identification field. But, iris identification still exists some technical problems in large-scale use, which makes iris identification become one of the most challenging issues in pattern recognition.Iris identification system usually consists of three parts:iris image acquisition, image preprocessing (iris location, normalized), feature extraction and pattern matching (the certification and recognition). Systematic analysis and sorting the previous studies, this paper makes deeply research and improvement of iris recognition algorithm. The main work of this paper includes the following respects:(1)In the iris positioning, this paper puts forward a more rapid localization algorithm. For the inside localization, firstly, roughly locates the circle point and radius of the pupil used the binary and gray-level projection methods. Then calculates the minimum variance of the candidate circles and the rough edge points accurately to obtain the parameters of inside edge. For the case of multiple minimum variances, uses point-by-point detection to ask out the most accurate parameters of inside edge. For the outside localization, firstly detects outside edge used the canny operator, and then crops and denoises the iris image based on the experience values. Then determines the outside edge through the minimum variance between the candidate points and the boundary points. Finally, asks out the most relatively accurate boundary parameters used point-by-point detection. Compared with the traditional algorithm, this algorithm is simple, easy to realize, improves the positioning accuracy and avoids the blinding search so as to improve the accuracy of the overall iris identification. Experimental results verify the feasibility of the proposed algorithm.(2)In feature extraction, this paper introduces the concepts of rough sets theory. Using the coarse screening thought and data discretization method of rough sets to roughly classify iris images. Firstly, extracts iris characteristics though applying 3 layers wavelet decomposition on the normalized image. Then, through Kohonen network discretization generates the data tables meeting the requirements of rough sets, which can roughly classify the iris and avoid the blindness of the subsequent matching process and save the recognition time.(3)In encoding and matching, in order to improve the defects that wavelet transform can't sign local characteristics texture very well, this paper firstly selects the third layer decomposition coefficients of wavelet which contains the most abundant information and then encodes quantitatively. Next divides the normalized image into sub-blocks. For each sub-block, applies wavelet decomposition and takes the third intermediate frequency components to encode quantitatively. These codes together with the codes of the third layer decomposition coefficients composed of the coding template for pattern matching. Finally, calculate the templates'similarity through the hamming distance to recognize the iris.Experimental results show that compared with the traditional methods, the wavelet transform combineds with rough sets theory has significantly increased the accuracy of the iris recognition. Wavelet transform, which has excellent properties in time and frequency domains has been a powerful tool in signal processing and can not be replaceable. Rough sets, which has excellent properties in dealing with uncertainty questions can effectively improve the system's anti-interference performance.The iris identification algorithm this paper proposed obtains well recognition results in the matlab7.0 environment, simulating the test samples of CASIA iris database (version1.0).