【摘要】 针对轨道检测图像分割慢和识别不精确的问题,提出了一种经过中值滤波和均衡化后调整差分计算的边缘检测预处理算法。首先使用均值、中值、维纳和SNN等4种去噪方法分别处理同一幅添加椒盐和高斯噪声的图像;然后使用3×3、5×5、7×7、9×9等4个模板分别对图像均衡化;最后在对Sobel增添45°、135°、-135°和-45°等四个梯度算子的基础上,求和算子内的计算点后均分255位的像素,再差分计算,并通过公式推导对比调整差分算子前后的抗噪能力。结果表明:(1)经过中值滤波的图像峰值信噪比最大,处理时间最短;(2)经过3×3增强后的图像PSNR最大;(3)经过该方法预处理后的轨道图像,峰值信噪比较大,原有轮廓信息保留度较高,能为后续快速分割和精确识别奠定基础。更多还原

【Abstract】 A preprocessing algorithm of edge detection based on median filtering and equalization and adjustment of differential calculation is proposed to solve the problem of slow segmentation and inaccurate recognition of track image detection. Firstly, the same image with salt and Gaussian noise is processed by four denoising methods such as mean, median, Wiener and SNN; then, the image is equalized by using four templates of 3 × 3, 5 × 5, 7 × 7, 9 × 9 respectively; finally, On the basis of adding four gradient operators to Sobel, such as 45°, 135°,-135° and-45°, the calculation points in the summation operator are equally divided into 255-bit pixels, and then calculate difference,and the Anti-noise ability before and after adjusting the difference operator is deduced by formula. The results show that:(1) The peak signal-to-noise ratio(PSNR) of the image after median filtering is the largest and the processing time is the shortest;(2) The PSNR of the image after 3 × 3 enhancement is the largest;(3) The track image preprocessed by this method has a high PSNR and a high retention of the original contour information, which can lay a foundation for subsequent fast segmentation and accurate recognition.更多还原