Image data enhancement method and system based on multi-template image

• Main Authors: LIU HONG, ZHENG YUANJIE, LI XINMENG, JIANG YANYUN, ZHANG JICHANG
• Format: Patent
• Language: Chinese; English
• Published: 06.08.2021
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; HANDLING RECORD CARRIERS; IMAGE DATA PROCESSING OR GENERATION, IN GENERAL; PHYSICS; PRESENTATION OF DATA; RECOGNITION OF DATA; RECORD CARRIERS

The invention provides an image data enhancement method and system based on multi-template images, and the method comprises the steps: classifying to-be-enhanced image data, and selecting N images from each type of image data as template images; performing pixel-level fusion on the template image to obtain a fused image; respectively inputting the fused image into a segmentation network and an evaluation network to obtain an initial probability distribution matrix and N initial confusion matrixes, wherein the segmentation network is coupled with the evaluation network; multiplying N initial confusion matrixes by the initial probability distribution matrix element by element to obtain N new probability distribution matrixes; respectively calculating cross entropies of the N new probability distribution matrixes and the corresponding template images, and forming a loss function with traces of the N confusion matrixes; and iteratively updating parameters of the segmentation network and the evaluation network bas