DR Image Enhancement Method for Complex Components Based on Zero-Reference Pixel Mapping Curve

DR (Digital Real-Time Imaging) technology has become the preferred method for industrial defect detection due to its fast detection speed and low radiation dose, and the enhancement technology for DR images of complex components will serve as a crucial foundation for ensuring effective defect identification. However, the current manual parameter adjustment method for enhancing DR images often results in inconsistent and inefficient outcomes. Additionally, existing deep learning-based enhancement methods face challenges such as the scarcity of paired datasets or high-quality reference images when applied to industrial DR image enhancement tasks. To solve the above problems, a DR image enhancement method based on zero reference pixel mapping curve is proposed. Firstly, the image pixel feature extraction network (IPFE-Net) was constructed to extract the features of the input image and obtain its pixel feature matrix (MOPF). Then, using MOPF as the parameter matrix, the input low-quality image is iteratively transformed according to the preset pixel transformation function to achieve pixel by pixel nonlinear grayscale mapping. Thirdly, a zero reference image loss function is designed to drive model training and constrain various physical features of the output image to achieve image enhancement. Through validation on actual enterprise datasets, the results indicate that this method effectively enhances the image quality and defect detection performance of DR images for complex components.