摘要:人耳检测作为人耳识别的关键步骤,其效果直接影响着人耳识别的性能。利用传统的Adaboost算法进行人耳检测,会出现的样本训练时间过长、过于依赖样本质量等问题。为克服这些问题,在训练不足及初始人耳定位不好的情况下,本文引进YCbCr肤色模型和多模板匹配技术策略对人耳进行精确定位。实验表明,改进后的人耳检测性能得到较大的提高,对动静态人耳均能达到准确定位和检测的效果,消除了部分误检和减轻了检测过程中出现的多框及人耳框选不全等现象,算法的鲁棒性较好。 关键词:Adaboost算法,肤色,多模板匹配 Combining Skin-color model and Multi-template Matching for Enhanced Adaboost Ear Detection
Abstract: As a key step of ear recognition, The ear detection impact directly on the recognition performance. But many problems will be caused by ear detection with traditional algorithm of Adaboost, Such as the long time of training、dependenting on the quality of samples overly. In order to e the problems, we introduce the model of YCbCr and the strategies of multi-template matching to perform ear location accurately under the conditions of insufficient training with Adaboost and the bad position of raw detection. The method that proposed eliminated some false detection of skin-color and the multi-location or the phenomena of plete selection. The experiments of practical detection show the performance of the proposed method is kept well and robust sufficiently whether under the static or the dynamic environments for ear detection. And the robustness of the algorithm is better. Keywords: Adaboost algorithm, Skin color, Multi-template matching. 0 引言 人耳识别技术是20世纪90年代末开始兴起的一种新颖的生物特征识别技术。研究表明,人耳具有唯一性、普遍性和稳定
