An image segmentation method based on appearance dictionary learning and shape sparse representation

Abstract

The invention discloses an image segmentation method based on appearance dictionary learning and shape sparse representation. The method comprises the following steps: moving an average grid to an initial central position; positioning the position of each mark point on the grid along the normal direction of the average grid; obtaining an initial segmentation result in combination with a shape sparse representation algorithm; using an algorithm combining appearance dictionary learning and normal searching to adjust mark points nearby the obtained initial segmentation result; performing constraint reconstruction on the adjusted grid by using a shape sparse representation algorithm again; adjusting the mark point again according to the reconstruction result in combination with the characteristics of the gradient vector flow field and the probability value of label reconstruction of appearance dictionary learning; and a final segmentation result is obtained by using a shape sparse representation algorithm. According to the method, the appearance prior information with the resolution capability and the shape prior information with the reconstruction capability of shape sparse representation are learned by fully utilizing the appearance dictionary, so that the mark point positioning algorithm and the sparse shape representation algorithm are complementary, and finally a more accuratelow-dose CT segmentation result is obtained.

Description

technical field [0001] The invention belongs to the field of image processing, and uses a locally distinguishable appearance dictionary learning and a shape sparse representation algorithm of a feature space to perform an accurate segmentation method. Background technique [0002] In the past 50 years, the incidence of lung disease in the world has increased significantly. According to statistics, in some countries in Europe and the United States and in large cities in my country, the incidence of lung tumors has ranked first among all male tumors. Low-dose CT images have been widely used in the diagnosis of lung diseases, especially in the analysis of lung tumors. Accurate and automatic lung segmentation can save doctors a lot of time spent on labeling lung structures. In addition, lung segmentation can improve the diagnostic accuracy of lung tumors and lung nodules by 17%. Therefore, automatic lung segmentation has aroused great interest of clinical experts. Shape and a...

AI Technical Summary

Problems solved by technology

[0003] However, many existing lung segmentation methods at home and abroad are based on a single prior (appearance or shape), such as: deep learning, graph cut and other methods, which are not very accurate and reliable for organ segmentation.

In addition, more segmentation algorithms tend to deal with normal data, and it is difficult to deal with abnormal data (disease or shooting technical defects, etc.)

Images