基于小波钝化的嵌入式图像处理算法研究

注塑模具保护系统普遍采用模板匹配图像处理算法,其存在实时性不强、精确度不高,以及对运行环境要求严苛等问题。论文提出了基于小波钝化的嵌入式模具保护图像处理算法,该算法运行于嵌入式平台,采用小波钝化的方法凸显待测残留物,无需进行模板匹配、图像配准和较正;提出了基于像素值统计的支持向量机检测算法,以适应嵌入式平台内存小的特点;引入支持向量机分类,有效解决了图像偏移带来的误差问题。MATLAB实验测得,该算法的模腔残留物检测平均准确率为85.71%,残留物检出平均耗时0.910s。结果表明,采用小波钝化和支持向量机的嵌入式图像处理算法,无论在算法检测精度还是算法响应速度方面均优于以灰度共生矩阵匹配算法和差影法为代表的图像匹配算法。

Embedded image processing algorithm based on wavelet passivation

Injection mold protection systems generally use template matching image processing algorithms, which have shortcomings of poor real-time, low accuracy, and stringent requirement of operating environment. This paper presents an embedded mold protection image processing algorithm based on wavelet passivation. This algorithm runs on an embedded platform and highlights residues using the wavelet passivation method without template matching, image registration and correction. A SVM(Support Vector Machine) detection algorithm based on the statistic of pixel values is proposed to make it adapt to the small memory of the embedded platform. The SVM classification is introduced as an effective solution to effectively solve the error problem caused by image shift. Experimental results by MATLAB show that for residues detection, the proposed algorithm achieves an average accuracy rate of 85.71% and average detection time of 0.910 s. The results also show that this embedded image processing algorithm using wavelet passivation and SVM outperforms other image template algorithms represented by gray level co-occurrence matrix algorithms and difference image algorithms in terms of both detection accuracy and response speed.