【作者】 侯明玉；

【导师】 胡正平； 田占海；

【作者基本信息】 燕山大学 ， 电子与通信工程， 2013， 硕士

【摘要】 目标检测是图像模式识别的首要前提，在自动控制、人机交互等领域有着重要的应用。本文对LBP进行深入的分析与研究，针对LBP变化单一和目标识别率低的缺点，设计一种基于多模块自适应局部二进制模式(δ-LBP)的目标检测方法。实现从传统局部二进制模式到多模块自适应局部二进制模式的改进，提高对目标的识别率。本文首先对图像进行预处理，然后计算图像中每个像素点与其局部邻域点的灰度值，通过选择自适应值δth形成δ-LBP，采用金字塔多模块δ-LBP直方图向量进行目标特征提取，最后由滑动窗相关运算得到检测结果，从而实现目标检测。经实验结果显示，基于金字塔多模块的δ-LBP描述的目标检测方法对方向、背景、距离、表情等变化都具有较好的鲁棒性，在机器视觉学习中具有进一步的研究价值。更多还原

【Abstract】 Targets detection is the precondition of image pattern recognition, which is important for cybernation and human-computer interactive. This thesis makes a deep analysis and research on LBP. In order to overcome the limits of LBP and the low rate of targets detection, a method of targets detection is proposed, which is based on multi-block self-adapt local binary pattern(δ-LBP). With the concept of optimization, traditional local binary pattern has been developed into Pyramid multi-block self-adapt local binary pattern, and at the same time the rate of identification.In this thesis, images are pre-process firstly, the gray-scale difference is calculated between each pixel and its local neighborhoods of an image. Then different self-adapt values are chosen to code the foregoing gray-scale difference. Secondly. Histogram vectors extracted from multi-block are used to describe the targets. Finally. Targets detection will be completed by sliding window to get the matching results. Experimental results show that the proposed method is robust to expression、background and distance variations which makes it valuable in the further research for machine learning.更多还原