We proposed a scheme for adaptively selecting filter parameters for detecting defects in various image textures. To implement the proposed scheme on a target steel image, we used wavelet reconstruction method. The adaptive parameter-selecting scheme was presented by analyzing the textures in an image and obtaining the appropriate parameters from a pretrained neural network by inputting these texture features. Experiments were conducted to detect corner cracks in the images of a steel billet, and the proposed scheme was compared with a conventional wavelet reconstruction method. The experimental results showed that our proposed scheme was effective in detecting defects in various textures of the target images.

我们提出了一种自适应选择滤波器参数以检测各种图像纹理中的缺陷的方案。为了在目标钢图像上实施该方案，我们使用了小波重构方法。通过分析图像中的纹理并通过输入这些纹理特征从预训练的神经网络中获取适当的参数，提出了一种自适应参数选择方案。进行了实验以检测钢坯图像中的拐角裂纹，并将所提出的方案与常规的小波重构方法进行了比较。实验结果表明，我们提出的方案可以有效地检测目标图像各种纹理中的缺陷。