基于改进八方向Sobel算子的图像轮廓提取方法

Image contour extraction method based on improved eight-direction Sobel operator

针对传统八方向Sobel算子的边缘提取算法对图像的检测效果差、漏检率高等问题,提出一种改进的八方向Sobel算子的图像轮廓提取方法。该方法首先采用高斯滤波算法对图像进行预处理,以减少图像噪声的影响,然后利用八个方向的模板计算各像素点的梯度,并将由各个方向模板提取到的各方向轮廓图像进行加权融合,以获得多个方向上的轮廓信息,最后采用区域生长算法进行后处理,以补充被漏检的边缘信息,从而使得提取到的图像轮廓更精确完整。通过在PyCharm平台上对CCPD数据集中的图片进行实验可知,由该算法提取到的轮廓图像的峰值信噪比(PSNR)为97.45、均方误差(MSE)为27.89和结构相似性(SSIM)为0.231,分别比传统八方向Sobel算子的边缘提取算法提高了54.4%,43.9%和92.2%。实验结果说明,文中改进算法提取的图像轮廓更精确,漏检率更低,整体检测效果更优。

In order to solve the problems of poor detection effect and high missed detection rate of the traditional edge extraction algorithm based on eight-direction Sobel operator,an image contour extraction method based on the improved eightdirection Sobel operator is proposed. In the method,the Gaussian filtering algorithm is used to preprocess the image to reduce the influence of image noise first,and then the gradient of each pixel is calculated by using the templates in eight directions,and the contour images extracted by the template in each direction are weighted and fused to obtain the contour information in multiple directions. The region growth algorithm is used for post-processing to supplement the missed edge information,so as to make the extracted image contour more accurate and complete. The results of the experiments on PyCharm platform for the images in CCPD data set indicate that the peak signal-to-noise ratio(PSNR)of the contour image extracted by this algorithm is97.45,the mean square error(MSE) is 27.89 and the structural similarity(SSIM) is 0.231,which is 54.4%,43.9% and92.2% higher than those of the traditional edge extraction algorithm based on eight-direction Sobel operator respectively. It shows that the image contour extracted by the improved algorithm is more accurate,the missed detection rate is lower,and the overall detection effect is better.