Research And Implementation Of Palmprint Preprocessing Algorithm

Biometric Identification Technology has been widely used in people's daily life, especially for the security field with stringent personal identification. Palmprint which contains rich texture information, with unique and steady features, has become a research hot pot in the biometric identification field. With the introduction of wavelet analysis and neural network, multi-resolution analysis to palmprint images, training and automatically learning to the extracted palmprint texture information are possible. The correct rate of palmprint recognition is continuously improved, which accelerates the market-oriented process of palmprint recognition.This paper anatomizes the preprocessing of palmprint recognition, especially for the research of palmprint contour and key point extraction, and proposes new algorithms.At the stage of palmprint contour extraction, this paper proposes the "the longest the most possible" algorithm to extract palmprint contour in two phases.This algorithm can extract single pixel and closed palmprint contour from palmprint images with many noises, and the noises in the palm won't be searched, so it has low time complexity.According to the palm morphology, this paper proposes a new key point extraction algorithm.The algorithm locates the region of key points, so it can shorten the key point search time. Because the algorithm uses the circumference whose circle center is the diagonal intersection point of the palm boundary rectangle to determine the key point region, it has low time complexity. With different setting positions of palm images, the algorithm has the same time complexity and has a strong robustness for the rotating palm images.This paper optimizes and improves these important algorithms in the preprocessing stage, so the preprocessing algorithm has low time complexity and time consuming of a palm image is about 40ms.Key point location algorithm has a strong robustness, so the whole preprocessing algorithm has a strong adaptability for the palm location. This paper preprocesses 244 palm images which contain noises and are rotated, and the correct rate of cutting ROI images is above 98%.