Prior Knowledge For Coarse To Fine PCB Resistor Segmentation

Accurate segmentation of Printed Circuit Board (PCB) resistor is an essential step for its defect level assessment. Limited by the available of training data, we firstly apply the traditional level set method for PCB resistor segmentation task, whose basic energy function can barely segment boundaries of resistor from similar background regions. To solve this problem, an improved energy function considering prior knowledge of PCB resistor is specially designed for this type of segmentation task. First, a PCB resistor shape energy term is involved into basic energy function to obtain coarse segmentation in the early stage. Second, a local region energy term considering distribution of local gray level is added to the basic energy function to achieve fine segmentation. Considering the lack of public dataset, we collect a new public available dataset for evaluation. Experiments on the dataset verifies the effectiveness of our improved level set method, which outperforms previous methods by 10% to 30%.