基于改进去伪匹配SURF算法的静态手语字母识别

针对SURF(Speeded-Up Robust Features)特征点不稳定性导致的伪匹配问题以及距离特征判别兴趣点真伪的方法未考虑到图像尺度变化情况的匹配问题,本文结合深度图像提出一种改进的去伪匹配SURF算法用于静态手语识别。改进的去伪匹配算法采用手势轮廓面积加权的距离特征判别法,同时,采用角度特征判别法综合剔除错误匹配点。实验表明,手势轮廓面积对距离的加权有效解决尺度变化对匹配点去伪的影响,且本方法对光照、复杂背景、角度和尺度变化有较强鲁棒性,具有更高的识别率。

Static sign Language alphabet recognition based on improved excluded Pseudo-match SURF algorithm

For the Pseudo-matching problem caused by the instability of SURF （Speeded-Up Robust Features） feature points and the method of judging interest points by distance characteristic without considering the image matching problem of scale changes, an improved excluded Pseudo-match SURF algorithm combining depth image for static sign language recognition is proposed. The improved algorithm exerts distance characteristic discrimination method weighted with gestures outline area, and angle characteristic criterion to eliminate false matching points syn-thetically. Experiments show that the method of distance weighted with gestures outline area can solve the influence of scale changes on pseudo matching points elimination effectively. In addition, the proposed method has relatively strong robustness and higher recognition rate against illumination, complicated background, angle and scale changes.