Research On The Key Technology Of Retinal Image Splicing

As a key part of human eye imaging,the retina is composed of the pigment cortex and the retinal sensory layer,which is close to the inner surface of choroid.Since the lesions caused by many fundus diseases usually show above the fundus vessels,obtaining retinal images is usually one of the important links in the diagnosis of fundus diseases and other systemic diseases in clinical medicine.Due to the Angle of the fundus camera photo co.,LTD.,a single image of the area is small,it is difficult to obtain effective image information,therefore,in the field of medical research and clinical diagnostic medicine,need to many different point of view of retinal image stitching together,to gain a more complete retinal fundus blood vessel information.However,limited by technical conditions,the existing image mosaics software cannot synthesize relatively efficient and complete images.Therefore,under the existing basic conditions,this paper proposes an improved SIFT image registration algorithm based on segmentation feature matching and a dynamic weighted fusion method to improve the visual effect of retinal image mosaics and fusion.The matching algorithm is proposed based on SIFT algorithm and according to the segmentation algorithm in the data structure.Appropriate and complete retinal images were selected from the retinal database as the experimental objects,and the retinal images were divided into different block numbers for testing,and then the retinal image registration and retinal image fusion were carried out.Through experiments to compare and analyze the algorithm in this paper and the traditional algorithm,the experimental results show that the SIFT segmentation feature matching algorithm mentioned in this paper can not only achieve low time consumption and low memory occupancy but also can obtain more ideal matching results.This paper analyzes three common image fusion algorithms,including the traditional weighted fusion method,the weighted fusion method based on the side of the Mosaic line,and the weighted fusion method based on the original image content.To improve the traditional weighted fusion algorithm,and puts forward the dynamic weighting method,this method according to the coefficient of dynamic curve,take two pictures of the grey value of the same position as the basis of the judgment weight,can effectively avoid the fixed coefficient,leading to ignore the specific situation of the individual characteristics of the image,not to all feature points which can lead to errors,influence the image fusion effect.The results show that thealgorithm has a good fusion effect,the images are not significantly deformed,the fundus blood vessels in the two images are accurately spliced,there is no obvious deviation,and the splicing line can be obviously eliminated,the image brightness of the splicing area of the two images is consistent.On the basis of theoretical research and experiments,this paper preprocessed retinal images,extracted and matched feature features with traditional SIFT algorithm,and spliced and fused retinal images according to the results of image matching,and obtained the spliced images in line with the visual effect through experiments.