基于身体健康状况表征的人体掌部精细纹路和颜色提取算法研究

发布时间：2018-06-10 21:20

本文选题：健康表征 + 主线修复　；参考：《吉林大学》2016年博士论文

【摘要】：传统中医通过观察人们手掌不同位置出现的纹理、颜色和形状等特征来判断身体健康状况,诊断身体疾病,而中医诊病往往需要积累多年的经验,若能通过掌纹识别技术实现身体健康状况的自动判断,对人们及早诊治疾病,避免病情恶化具有重要意义。随着近10年生物识别技术的快速发展,掌纹识别技术作为生物识别家族的重要成员,已得到国际上的认可,并取得了丰硕的研究成果。然而应用现有的掌纹识别技术来判断身体健康状况仍然存在很多困难,如:(a)现有方法提取的掌纹主线浅且易断续,不利于后续处理;(b)现有方法极少可用于提取一些异常纹路;(c)手掌颜色能够反映整个身体健康状况,现有方法不能识别手掌上的颜色斑点等。本文针对这些困难,从表征身体健康状况这一背景出发,展开掌部精细纹路和颜色的提取方法研究,主要工作包括以下五个方面:(1)针对掌纹主线浅易断续,不利于后续处理等问题,本文提出了一种修复掌纹主线的新方法,即先基于图像的能量按照高斯函数膨胀,然后计算子图像和非亮点的概率分布,最后图像的像素按指数函数回归。对比传统解决纹线断续问题采用膨胀、细化的方法,本方法能够解决误连接和远距离不能连接的问题,从而更好的解决了远距离断续主线无法修复和主线误修复的问题。(2)现有方法极少提取掌纹中反映身体健康状况的精细纹路,本文提出了一种使用多光谱技术与NSCT变换相结合的掌纹纹路提取方法。由于NSCT变换不仅具有平移不变性,而且进行掌纹纹理分解时,NSCT变换有多尺度和各向异性的特点,有利于进行后续的掌纹图像融合。因此我们采用NSCT变换对各个光谱下的掌纹图像进行分解,并根据掌纹纹路特征对分解后的子图进行融合,最后用形态学方法细化并提取掌纹的精细纹路。实验结果证实,该方法提取的精细纹路丰富而且清晰。(3)针对已有算法提取“十”字纹和“米”字纹时,识别率不高的现象,本文应用霍夫变换的累加器投票原理对掌纹中的“十”字纹和“米”字纹进行提取。我们先用自己开发的多光谱掌纹平台采集掌纹,构建“十”字纹、“米”字纹数据库,然后使用新高斯预处理模型对图像进行处理,最后用霍夫变换提取“十”字纹和“米”字纹。实验结果表明我们所提方法能有效地提升识别率,为进一步进行掌部表征所对应的身体健康状况研究奠定了基础。(4)由于手掌颜色能够反映整个身体健康状况,如何准确而快速的识别手掌上的颜色斑点,是表征身体健康状况的关键。本文提出了一种H-S-Gray斑点提取算法,该算法在不改变图像对比度和尽可能减少颜色失真的情况下,通过颜色空间的拉伸、量化等预处理操作,使处理后的图像有利于颜色斑点的识别。(5)为了实际验证基于身体健康状况表征的掌部颜色及纹路提取算法的有效性,本文研制了一个多光谱掌纹采集平台。针对单一光谱下的掌纹图像信息量小且具有局限性从而造成识别精度低等问题,研制的多光谱掌纹采集平台能够采集六个光谱下的局部掌纹图像,得到手掌各个层次的信息,从而给识别带来更高的准确率。既可以实时观察掌纹图像,又可以选择性的采集并存储掌纹图像信息。综上所述,本文对身体健康状况的掌部精细纹路及颜色提取算法进行深入研究,为真正实现自动表征身体健康状况提供理论支持。
[Abstract]:Traditional Chinese medicine can judge the health condition of the body by observing the texture, color and shape of the palm of the palm of the hand to judge the health of the body and diagnose the diseases of the body, and the medical diagnosis of traditional Chinese medicine often needs to accumulate years of experience. If it can realize the automatic judgment of the health condition of the body through the palmprint recognition technology, the people can diagnose the disease early and avoid the evil of the disease. With the rapid development of biometric recognition technology in the past 10 years, palmprint recognition technology, as an important member of the biometric family, has been recognized internationally and has achieved fruitful research results. However, there are still many difficulties in the application of the existing palmprint identification technology to determine the health status of the body, such as: (a) The main thread extracted by the method is shallow and discontinuous and is not conducive to subsequent treatment; (b) the existing methods are rarely used to extract some abnormal patterns; (c) the palm color can reflect the health of the whole body, the existing methods can not identify the color spots on the palm of the hand. The main work includes the following five aspects: (1) a new method to repair the main line of the palmprint is proposed in this paper, which is based on the Gauss function, and then calculates the probability of the sub image and the non bright spot. The pixels of the final image are returned by exponential function. The traditional method of expansion and refinement is used to solve the traditional line breaking problem. This method can solve the problem of misconnection and remote connection, which can better solve the problem that the long distance intermittent mainline can not be repaired and the main line is mistaken. (2) the existing methods rarely extract the palmprint. In the fine lines reflecting the health of the body, this paper proposes a palmprint extraction method using the combination of multi spectral and NSCT transform. Because the NSCT transform not only has the invariance of translation, but also the texture decomposition of the palmprint, the NSCT transformation has the characteristics of multiscale and anisotropy, which is beneficial to the subsequent palmprint image fusion. So we use NSCT transform to decompose the palmprint images under each spectrum, and combine the decomposed subplots according to the palmprint features. Finally, the fine lines of palmprint are refined and extracted by morphological methods. The experimental results show that the fine lines extracted by this method are rich and clear. (3) the extraction of the existing algorithms is used to extract the existing algorithms. The recognition rate of ten "ten" and "rice" is not high. In this paper, the "ten" and "rice" lines in the palmprint are extracted by using the accumulator voting principle of Hof transform. We first use the multi spectral palmprint platform that we developed to collect the palmprint, construct the "ten" pattern, the "rice" pattern database, and then use the new Gauss. The preprocessing model deals with the image. Finally, the "ten" and "rice" lines are extracted by Hof transform. The experimental results show that the method can effectively improve the recognition rate and lay the foundation for further research on the health status of the body. (4) the color of the palm can reflect the health of the whole body. How to identify the color spots on the palm accurately and quickly is the key to the representation of the health of the body. In this paper, a H-S-Gray speckle extraction algorithm is proposed. The algorithm, which does not change the contrast of the image and reduces the distortion of the color as much as possible, makes the processed images by stretching, quantizing, and so on in the color space. In order to verify the recognition of color spots. (5) in order to verify the effectiveness of the palm color and pattern extraction algorithm based on the representation of physical health, a multi spectral palmprint acquisition platform has been developed in this paper. The spectrum palmprint acquisition platform can collect six local palmprint images under the spectrum, get information at all levels of the palm, and bring higher accuracy to recognition. It can observe the palmprint image in real time, and can selectively collect and store the palmprint image information. The color extraction algorithm is deeply researched to provide theoretical support for the automatic representation of physical health.
【学位授予单位】：吉林大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：TP391.41

【相似文献】