Research On Palmprint Recognition Algorithm Based On Local Descriptor

In recent years, with the development of network and informatization in society, there is higher demand for personal identity confirmation. Biometric identification technology, as a kind of important and reliable solution, attracts extensive attentions. By connecting computer with optical, acoustics, biological sensors, biological statistics and other high-tech means, biological recognition technology identifies personal confirmation by distinguishing the characteristics that inherent in human body. Because the characteristics of human body is unique, unrepeatable, non-reproducible and can’t be stolen or forgotten, biometric identification technology has high reliability.Compared with other features, palm print characteristic can be easily gotten. Its characteristics are obvious and steady, so it has great developmental potential and application space. Palm print recognition is also a kind of non invasive method, so the users are more likely to accept and the acquisition equipment requirement is not high. The information contained in palm print is much more abundant than that of a fingerprint, so we can confirm a person’s identity by his grain line features, point feature, texture features, geometric features that contained in palm print. Therefore, in theory, palm print has better resolution and higher identification ability than fingerprint. Due to these advantages, the palm print identification technology attracts more and more researchers’ attention, and is widely put into practical application.Palm print recognition system and key technology in palm print recognition were studied in this thesis, the research work and innovation points summarized as follows:(1) As for influence from the extraction of interested area of palm print image on recognition, the method of locating and segmenting based on the tangent fitting was adopted to extract the interested area of palm print. After extracting the contour line of the palm print, calculate the clearance between the index finger and middle finger, ring finger and little finger, and set up coordinate system based on the two common tangent points as the benchmark, and separate the region into several fixed size regions as interested areas. This method is efficient and accurate, and laid the groundwork for the back extraction of palm print features. (2) In order to achieve a higher classification performance with low cost, this paper proposes a palm print recognition based on SIFT descriptor and Bhattacharyya coefficient algorithm. This method firstly uses SIFT matching algorithm to get matching points, and then calculate the Bhattacharyya coefficient of10×10surrounding areas of matching points as features, at last use fuzzy neighbor classifier to classify. Experimental results show that the algorithm of palm print image rotation, scale and intensity change has robust and high precision characteristics.(3) In order to upgrade the precision and decrease computational cost, this paper proposes a new palm print recognition algorithm based on region invariant descriptor. Extract all the training samples of SURF belonging to the same class to each other, and then calculate every key point matching rate over1/2of the training sample, and the category database contains the mean value and the variance of SURF. In the identification phase, we utilize SURF to detect keypoints from the palmprint image to be identified, and then the fuzzy matching degrees for all palmprint patterns in the library are calculated. A fussy reasoning approach is developed for matching. Experimental results demonstrate that the proposed method yields a better performance in terms of the correct classification percentages compared with the recent palmprint recognition algorithms. It is also shown that the proposed approach is robust to the variations of orientation, position and illumination, and yields observably low computational cost.