The two-dimensional (2D) Gabor function has been recognized as a very useful tool in feature extraction of image, due to its optimal localization properties in both spatial and frequency domain. This paper presents a novel palmprint feature extraction method based on the statistics of decomposition coefficients of the Gabor wavelet transform. It is experimentally found that the magnitude coefficients of the Gabor wavelet transform within each subband uniformly to approximate the Lognormal distribution. Based on this fact, we create the palmprint representation using two simple statistics (mean and standard deviation) as feature components after applying the logarithmic transformation of Gabor filtered magnitude coefficients for each subband with different orientations and scales. The optimum setting of the number of Gabor filters and orientation of each Gabor filter is experimentally determined. For palmprint recognition, the popularly used Fisher Linear Discriminant (FLD) analysis is further applied on the constructed feature vectors to extract discriminative features and reduce dimensionality. All experiments are both executed over the CCD-based HongKong PolyU Palmprint Database of 7752 images and the scanner-based BJTU_PalmprintDB (V1.0) of 3460 images. The results demonstrate the effectiveness of the proposed palmprint representation in achieving the improved recognition performance.
The multiscale and multidirectional transform is a tool that has been used widely in the last decade for image processing. This paper presents a novel image feature descriptor for palmprint recognition based on the Dual-tree Complex Wavelet transform (DT-CWT), which provides a local multiscale description of images with good directional selectivity, invariance to shifts, insensitive to illumination and in-plane rotations. Instead of exploiting the DT-CWT-derived coefficients directly, which are highly-dimension, we investigate a statistical model to characterize the image in the transform domain. It is experimentally founded that the DT-CWT-derived magnitude of one palmprint image approximates a lognormal distribution, i.e. the logarithmic transformation of DT-CWT-derived magnitude is close to a Gaussian model. Thus the shape parameters (mean and standard deviation) of Gaussian are exploited to construct the feature descriptor for palmprint recognition in this paper. This process brings computational efficiency. For capturing the spatial structure information, each image is partitioned into many quadtree-based subblocks, whose DT-CWT-derived magnitude destributions are similar to that of the whole image. Finally the Fisher Linear Discriminant (FLD) classifier is used for palmprint recognition.Experiments are carried out on the BJTU_PalmprintDB (V1.0) of 3,460 images. The results demonstrate the high recognition performance of our proposed method.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.