With the increasing attention of public security and rapidly development of mobile internet technology, a number of different iris recognition systems are widely deployed in the real world due to the reliability of iris iometrics, which poses new challenges to iris recognition algorithms. In this paper, by adopting the strategy of “divide and conquer”, we focus on cross sensor iris recognition, and start with the traditional procedures, feature extraction, feature selection, feature mapping and fusion with periocular region. By using technologies of image processing and pattern recognition, factors of imaging sensors and environments will be diminished or eliminated,which is of great help to the deeper research and large-scale applications of iris recognition systems. The main contributions are as follows : 1. A robust feature extraction method for cross-sensor iris recognition is proposed. By analyzing the limitations of existing feature extraction methods, we propose a multi-direction weighting ordinal features(MultiOM) which can obtain robust iris features with better discriminative abilities. We also extend this feature extraction method to periocular feature extraction. In addition, a novel fusion strategy of iris texture and periocular texture is proposed. 2. We propose the coupled feature selection method for cross-sensor iris recognition. This cross-sensor matching problem is transformed to a framework of feature selection method based on l21-norm. However, this framework is difficult to solve because of the non-convex problem. Based on half-quadratic optimization, an iterative optimization algorithm is developed to obtain the global optimum. In addition, a similarity measure constraint is introduced to this framework, so as to select the more powerful features for cross-sensor iris recognition. 3. A margin based feature selection method for cross-sensor iris recognition via linear programming is proposed. To take advantage of the large margin principle and by means of novel constraints, relationship between coupled feature spaces is established. We reformulated the constraints in terms of linear inequalities. Then this model can be computed conveniently via the Simplex algorithm. 4. A feature mapping method is proposed for cross-sensor iris recognition. After the step of feature selection, these selected features from coupled feature spaces will be transformed to high-dimension subspace with better discriminative ability and smaller differ...
修改评论