Research On Medical Image Feature Detection And Automatic Mosaic Images

Age-related eye diseases will become more common and develop into a serious public health problem as the population shifts sharply towards the elderly.Fundus mosaic image is of great significance for diagnosis and intervention planning of surgery.However,a fundus image taken by fundus camera or scanning laser ophthalmoscope(SLO)can only capture the local area of the eye,so it is of great research to produce a clear mosaic image with wide field from multiple high-resolution images.Because of advantages such as low radiation and low price,X-ray image is still irreplaceable in the diagnosis of diseases.X-ray images,such as spine or lower limb images,can only be partially captured due to the limitations of hardware such as detectors.For diseases such as scoliosis and lower limb deformity,doctors need X-ray mosaic images to provide better clinical diagnosis.The research contents of this paper include:(1)Automatic stitching of fundus images based on features.In this algorithm,convolutional neural network is used to segment the vascular structure in the fundus image,and then the method based morphology operations is proposed to extract the bifurcations of the vascular structure as the features.Then,the image was matched,and the Random Sample Consensus(RANSAC)algorithm was used to purify the image matching pairs and calculate the image homography matrix to achieve image registration and alignment.All images to be stitched are transformed to the same coordinate system by using different projection transformation models.Finally,used the multi-band blending method to blend the image seamlessly to generate mosaic image.The data came from 14 pairs of different eyes,each taking 2-9 fundus images.A total of 62 images,190 pairs of overlapping images.Root mean square error and matching recall rate are statistical metrics to measure matching performance.The root-mean-square error obtained by the algorithm in this paper is low,and the matching recall rate is higher than other algorithms.Image stitching software was chose compared with the proposed method.The stitching softwares include AutoStitch,phtoshop cs6 and ICE(Image Composite Editor).Experimental results show that the algorithm can achieve accurate matching and obtain high-quality mosaic images.(2)Automatic stitching of X-ray images based on region.This paper proposes a stitched algorithm based on template matching.First,the image is preprocessed,and then the gray level information of the image is accurately calculated by the template matching method to obtain the translation parameters of the stitched images,determine the overlapping area between the images,and achieve the image registration.Finally the multi-band blending method is used to blend the image to generate mosaic image.In the experiment,the normalized cross correlation number was used as the quantitative evaluation metrics to analyze the relationship between the width of the template and the normalized correlation coefficient,and the most appropriate width of the template(50%)was chose.In low limb images,the average of the normalized correlation'coefficient reached 98%,and in spine images,the average of the normalized correlation coefficient was 85%.Finally,mosaic images of the spine and low limb were stitched respectively,and the experiment showed that the mosaic image is excellent.