Image segmentation method and evaluation method and image fusion method thereof

Abstract

The invention discloses an image segmentation method. The method comprises: step one, a historically stored image is divided with K view angles and historical clustering centers are obtained by a classical FCM algorithm; step two, when a new noisy image is processed, a to-be-processed image is divided into K view angles, a clustering center, obtained at the step one, of a related historical similar image is fused based on the classical FCM algorithm;, and a new objective function of a multi-view-angle FCM algorithm with introduction of a transfer learning mechanism is constructed; and step three, a final membership degree matrix of a currently processed image is obtained according to membership degree matrixes of all view-angle images of the current images and view-angle weight vectors W, defuzzification is carried out to obtain a space division result of the current image. According to the image segmentation method, a multi-view-angle fuzzy clustering image segmentation method having the transfer learning capability is employed; and on the basis of the transfer capability, multi-view-angle coordinated image segmentation is realized, so that the segmentation precision is improved.

Description

technical field [0001] The present invention relates to the technical field of image processing, and more specifically, the present invention relates to an image segmentation method, an evaluation method thereof, and an image fusion method. Background technique [0002] The specific performance of transfer learning is that when using this theory to build a model, it will consider the existing similar models in the past, use the previous model as a reference body, and then combine the current environment for modeling. This new modeling method will greatly improve The early modeling efficiency and the effective and reasonable use of historical reserves also contribute to the initial stability of the model. Compared with the traditional modeling method that does not consider similar historical scenarios and only considers current scenarios, all starting from "zero", This strategy is faster and more effective. [0003] Fuzzy C-means algorithm (Fuzzy C-means, referred to as FCM)...

AI Technical Summary

Problems solved by technology

[0004] Chinese invention patent application 201210384176.2 provides a fuzzy clustering image segmentation method with transfer learning ability, which uses a single perspective to segment the image. During the segmentation process, the segmentation accuracy may not be high, and ideal segmentation results cannot be obtained. It is even more unable to handle the segmentation of multi-dimensional images

[0005] However, traditional image segmentation evaluation indicators often treat the target pixels and background pixels in the segmentation results in the same way, ignoring the importance of pixels at different positions, which is unreasonable in practical applications.

Images