基于跨尺度特征融合的内窥镜图像增强算法OA

The clinical endoscopic image often suffers from low-quality imaging due to uneven supplementary light sources and reflections from human tissue mucus,resulting in poor image quality,for instance,a large quantity of overexposure.However,the current deep learning based image enhancement algorithms have low feature extraction capabilities due to fixed-size feature fusion,which leads to poor enhancement effects.Therefore,an endoscopic image enhancement algorithm based on cross-scale feature fusion is proposed.In the algorithm,a convolution module(CM)is constructed for high-performance feature extraction and a spatial pyramid pooling-fast(SPPF)module is used to realize the pooling operation of feature maps with different scales.Additionally,a cross-scale feature fusion(CFF)module is introduced into different scales of network layers to achieve multi-scale feature fusion and context information propagation,so as to improve image detail capture and image quality.Experimental results show that the proposed algorithm outperforms the existing algorithms in PSNR and SSIM,in which the PSNR is improved by 9.9%,and the SSIM by 15.4%,achieving high-quality endoscopic image enhancement.