摘要： A filter bank (FB) is an integral part of any image processing system. The designing of a FB generally involves modifying an existing FB or focusing on a particular property of the filter bank. Such FBs limit their use to a particular image. Through our work, we have devised a unique and novel approach for designing a two-dimensional arbitrary shape filter bank (2-D ASFB). This FB is inherently 2-D and eliminates the need for transforming a one-dimensional FB into 2-D. Its arbitrary nature expands its application to any image as compared to regular-shaped FBs currently in use. The novelty of the design lies in the fact that the designed FB can match the frequency spectrum of any image by reducing the error function between the frequency spectrum and the desired filter response of the FB. The error function has been minimized using the eigenfilter approach. After designing the low-pass analysis filter, perfect reconstruction constraint has been used to get a low-pass synthesis filter. In this paper, we have demonstrated the use of the 2-D ASFB specifically for contact lens detection (CLD). The proposed CLD system focuses on feature extraction using the 2-D ASFB. The support vector machine classifier is the same as in the existing systems. The results show improved correct classification rate as compared to the existing systems for IIITD and ND2013 contact lens database. This 2-D ASFB overcomes limitations posed by the existing filter banks with respect to separability, directionality, orthogonality, and shape. This FB can be effectively applied to any feature extraction application such as pattern recognition, biometrics, medical image processing.