| The acquisition of iris images is the first step of iris recognition, as the result, if the iris acquisition system can get clear iris images, the efficiency and accuracy of the iris recognition algorithm will be greatly improved. The acquisition of iris images will be different because of different equipment, and in general there are two methods: 1. one iris image is got at one time; 2. the acquisition is conducted as video stream, in other words, several iris images are got during given period, then one or several images satisfying the need of quality are selected to be used for recognition. In our system, the acquisition equipment belongs to the second method, so a kind of good quality evaluation method of iris sequence images is needed, which should compare the quality of the iris images sequence and justify whether the quality satisfies the recognition need.During the evaluation of iris images quality, it is an important method to make the result of objective evaluation methods to be coincident to the one of subjective evaluation methods by use of human visual system property. Human visual system has some basic properties. Looking from visual psychology, the vision is a kind of active reception activity, and it not only correlates to biological factor, but also is greatly affected by psychological factors. When human appreciate the image, they will unintentionally have some interest in some sections in the image, and these sections are called "region of interest". In general the visual quality of the total image is determined by the quality of ROI, and the quality of "region of un-interest" has little effect on the quality of the total image. During the quality evaluation of iris images, it is self-evident that the iris texture circle section is our ROI, and its quality determines the quality of the total image. Considering the practical need of iris recognition, during the iris sequence image quality evaluation we select the follows factors as the key points:a. The area of the iris circle section and its ratio to the one of the pupilIn general, the bigger the acquired iris circle section's area, the better it will be beneficial to the recognition, however, if the ratio of the iris circle section area to the pupil area is too big, then it is owing to the acquisition deformation, and it will deform the iris image. Consequently, a good iris image should be the one whose iris circle section area is suitable in the total image and its ratio to the pupil area is also suitable, which can be checked by the iris locating result.b.The detail richness level of the iris circle sectionThe detail richness level of the image texture can be described by some well known texture properties. Suppose the image belongs to the Gauss model, if the mean energy of a section is bigger than a given threshold value, then this section can be considered as texture section. The bigger the summed mean energy of all the texture sections in an image, the richer the detail of the image.The method to locate the iris includes: the image edges are detected by Canny edge detection operator; the normal line algorithm is used to coarsely estimate the pupil center and iris inner circle radius; suppose the inner and outer circles of the iris have the same center, Hough transform is used to locate the center and get the radius of the iris inner and outer circles, r and R respectively, then the area of the iris circle section and its ratio to the one of the pupil can be used as a quality evaluation factor.Wavelet analysis is a kind of analysis method combining time and frequency fields, which can decompose the signal in time and frequency fields at the same time, and the consequent multi-resolution data is apt to consider the human visual characteristics. Mallat proposed a tower algorithm to get the wavelets, which makes the discrete wavelet transform appear as digital filter groups. For the 2-D signal source such as the still image, the 2-D filter is needed to process it. Considering the separability of the wavelet...
|
PDF Full Text Request