An Endoscopic Image Super Resolution Method Based On Residual Network And Mixed Attention Mechanism

With the continuous development of medical equipment,the use of endoscopic images to determine the progression of a disease has become one of the most common means.The higher the quality of the endoscope image,the more detailed the image,the more helpful it is to the doctor’s diagnosis.There are two methods to improve the resolution of endoscope image: improving the hardware equipment and image reconstruction algorithm of super resolution.However,there are many limitations on the cost and technology of hardware equipment improvement,and it is not practical to improve the image quality.Therefore,the research on the endoscope image super-resolution reconstruction algorithm has important practical significance.Current endoscopic image super-resolution methods mainly use traditional methods and deep learning based methods.For the past few years,image super resolution reconstruction algorithms based on deep learning show good advantages,but still face some problems such as unclear image composition and difficult convergence in training.To address the above issues,this paper proposes a residual mixed attention mechanism for super-resolution of endoscopic images,which effectively improves the resolution of endoscopic images.By introducing the residual network,the problem of gradient disappearing caused by the deepening network in the course of network training is avoided to a great extent,and the problem of network degradation is effectively solved.By introducing the mixed attention mechanism,the feature map can combine the attention weights of channel and space,which is conducive to extracting more effective features and high-frequency information,making the reconstructed image clearer and richer in detail.So as to verify the validity of the algorithm,this paper conducts experiments on a public dataset and a real endoscopic image dataset.The experimental results showed that the method proposed in this paper was applied to endoscopic images with different departments and different magnifications,and the PSNR and SSIM were the highest.Moreover,by observing and comparing the local detail maps of different methods,The method proposed in this paper reconstructs the microscopic blood vessels and fascia layers in the endoscopic images more clearly.