Panoramic Imaging Method Based On Sequence Images Of Digestive Endoscopy

Endoscopy is an important means of in vivo diagnosis and clinical practice of minimally invasive surgery,doctors can observe the patient’s body with the help of an endoscope to obtain information on the patient’s internal tissues and organs,including the lesion,the type and size of the lesion,and so on.However,due to the narrow field of view of the endoscope,it brings certain difficulties to the clinical practice of minimally invasive surgery,especially when it is necessary to grasp the whole picture of the lesions in the tissues and organs,and quickly locate the lesions and make corresponding preoperative planning,unfavorable for actual surgical operation.3D panoramic images with a large field of view can solve the problem of narrow field of view,although two-dimensional panoramic stitching can also increase the field of view,it loses the depth information in the three-dimensional information,this is not conducive for doctors to grasp the spatial structure information of tissues and organs.The three-dimensional panoramic image with depth information and texture has great application potential in assisting doctors in clinical practice.In this paper,the three-dimensional panoramic imaging of organs based on sequence images of digestive endoscopy is studied.The main contents are as follows:（1）The algorithm for feature extraction and description of endoscopic images with weak texture is studied,and an algorithm for feature extraction and description of endoscopic images based on deep learning is proposed,and use deep learning method to complete the feature extraction and description of endoscopic images,it solves the problem that the traditional feature extraction algorithm cannot extract enough feature points on the endoscopic image with weak texture.Then,according to the matching of the endoscopic image features,the polar constraint relationship is established,and then the essential matrix is obtained,and finally the camera pose is obtained through the singular value decomposition of the essential matrix.（2）This paper studies the imaging process of the camera and occlusion culling,and then constructs a virtual camera,combined with the camera pose,the virtual camera was used to observe the reconstructed 3D contour model of the organ based on the CT sequence images,and the visible model triangle faces were determined for each image.（3）After obtaining the registration relationship between each image and the model,this paper studies the texture mapping method of the 3D contour model,and the texture mapping method based on Markov random field is improved.After obtaining the texture mapping results of the 3D contour model,global and local texture seam processing is performed on the texture mapping results respectively.First,global color consistency adjustment is performed on the texture mapping results of the 3D contour model,the mapping results are then further processed locally using Poisson gradient fusion.The experimental results show that the proposed method can obtain 3D panoramic images with better visual effect.