Research on Image Alignment and Image Processing Based on Deep Learning

Image alignment is a typical problem and technical difficulty in image processing, and using reasonable image alignment algorithms can accurately integrate multimodal image information into the same image. In this paper, the alignment is mainly divided into three stages, i.e., image preprocessing, alignment algorithm to realize accurate alignment, recognition and quantitative evaluation. First, the tomographic image needs to be preprocessed. In order to simplify the amount of subsequent operations, the original image is binarized; the improved Gaussian filter is applied to edge denoising smoothing and the Laplace operator combined with intensity weights is used to perform image edge enhancement for subsequent recognition and alignment; the Canny algorithm is used for edge detection, and the double threshold algorithm is used to detect and connect the edges, and the region is segmented using a diagnostic method based on the depth of the semantic features to establish and solved a hybrid weighted loss function, and a vector machine classifier based on wavelet kernel space is utilized to achieve detection and segmentation. This paper mainly uses some existing optimization algorithms for image preprocessing, alignment, and recognition segmentation, and through model building solving and theoretical analysis, it is concluded that the three- phase model of this paper has certain feasibility for image alignment.