Quality Evaluation And Segmentation For Iris Image

The accurate location, time consumption and quality of the iris images affected thepromotion and application of the iris recognition system. In order to ensure the quality ofiris image before location, the system need to add image quality evaluation. In this way, thesystem efficiency can be improved. Therefore, the paper included the following contents:1) The system of iris image quality evaluation was established before segmentation.Doing so can improve the efficiency of the system. According to the characteristics of theacquired iris images, in this paper, five indexes of quality evaluation was proposedsequentially. The five indexes were eyelid occlusion detection, living detection, definitiondetection, eyelashes occlusion detection and cosmetic contact lens detection, respectively.Among them, the gray-level distribution of the iris image was used to detect eyelidocclusion and cosmetic contact lens. The difference is that gray-level distribution followingremoving the coarse location of the pupil was used to detect cosmetic contact lens. Thesub-band energy information after wavelet packet decomposition was used to detectdefinition and eyelashes occlusion. The difference is that multi-dimensional wavelet packetdecomposition was used to detect eyelashes occlusion. Living detection was detectedaccording to the contractile properties of the pupil. Experimental results showed that irisimage of quality could be accurately detected if the system of image quality evaluation wasattached. At the same time, the time-consuming could be shortened.2) To enhance the efficiency of location, spots were needed to be removed. In thispaper, the median filtering, image enhancement and binarization were used in removingspots. These methods can remove most of the spots in the iris image. And then the spotswere completely removed through the coarse location and the mean of gray value. Thesemethods were simple and effective. They can remove the spots in all of the iris images,without affecting the image edges.3) Two methods were proposed about iris image location and segmentation in thispaper. They are respectively regional analysis and pulse coupled neural network. In themethod of regional analysis, the regions were selected in four directions based on the coarse location of the inner boundary. And the edge points on the boundary of iris inner can bedetermined through the analysis of the gray gradient in the selected regions. Then someeffective regions were determined based on the precise inner boundary. The outer boundarypoints were got by analyzing the edge detection of the effective regions. Finally, the innerand outer boundary of iris were obtained by Hough transform. This method had a lesscomplexity, so it can help improve efficiency. The iris located and segmented by pulsecoupled neural networks can make it more conform to meet the requirement of the humanvisual on the image segmentation. The edge points on the boundary of iris were determinedby simplifying the improved PCNN model. In this PCNN model, the terminal time was gotbased on minimum cross entropy. Also, the Hough transform was used to determine theboundaries of iris. Experimental results showed that PCNN could improve the accuracy ofiris location and segmentation. Because the search range of Hough transform was reduced,so it could reduces the time consumption.4) After location and segmentation, the iris image was normalized, encoded andcalculated the similarity. Firstly, the Rubber Sheet Model was used to normalize the image.The2D-Garbor filters were used for extracting texture features. The filter results of positiveor negative information was compiled as gray code. The similarity was calculated byhamming distance. Large numbers of iris images from the library of the CASIA-Iris wereused in the experiments, so that the false acceptance rate and false rejection rate can beacquired. Experimental results showed that the locating method in the paper can effectivelyimprove the accuracy of the system.