基于中值滤波和Hu矩向量的手语识别

Sign language recognition based on Hu invariant moments and euclidean distance

为提高手语图像的正识率,提出一种基于中值滤波和Hu不变矩特征量相结合的手语识别方法。对待匹配图像进行预处理,计算图像的7Hu矩特征量,通过欧氏距离计算与标准手语特征向量的相似程度,识别手语。对旋转、缩放和平移等图像的非本质性改变具有很好的鲁棒性,克服了手语图像采集中光照不均的影响和传统手语识别中基于图像分割造成的边缘信息损失。实验结果表明了该方法对手语识别的有效性,特别是对含有椒盐噪声信噪比较高的图像具有很好的效果。

To increase accuracy rate of sign language recognition, a recognition method based on Hu invariant moments and euclidean distance of sign language image is introduced. Firstly, its 7Hu moments are calculated, the recognition of 30 letter gestures is performed by computing distance between matching and template image to recognize image. The algorithm remains constant when the image rotates, transfers and scales. The method overcomes the impact of illumination in capturing image and marginal information loss caused by image segmentation. The experiments show the validity of this method especitaly a successful result of images with a high rate of pepper-and- salt noise.