Image Segmentation with Multi-Scale Feature Fusion of Local Binary Patterns for Flexible Integrated Circuit Packaging Substrates

Image segmentation is employed to address the problem of recognizing the Region of Interest (ROI) in detecting defects. However, detecting defects in Flexible Integrated Circuit Packaging Substrates (FICS) has several challenges, including low brightness, diverse defect types, and defect textures that resemble the substrate textures. Existing image segmentation methods fail to distinguish the ROI region from the substrate texture, leading to misjudgments during detection. This paper introduces a method for Image segmentation using a multi-scale feature fusion technique based on Local Binary Patterns (LBP). Firstly, the image undergoes preprocessing using the histogram equalization theory, significantly enhancing its contrast. Next, the multi-scale feature fusion technology based on LBP is utilized to obtain the multi-scale feature saliency map. Lastly, a morphological closing operation is performed on the image to fill any holes within the circuit and eliminate small target areas that are below the predefined threshold. The results demonstrate that the proposed method in this study achieves higher segmentation performance, compared to the threshold method, region method, and edge method.