Gastrointestinal endoscope image anomaly detection method based on local feature and class label embedding constraint dictionary learning

Abstract

The invention discloses a gastrointestinal endoscope image anomaly detection method based on local feature and class label embedding constraint dictionary learning. Firstly, an original image is preprocessed; eXTRACTING REGIONS OF INTEREST, extracting color and texture features from the image data and fusing the color and texture features; a test set and a training set are constructed; establishing a dictionary learning model using an atomic class label embedding item and a Profiles structural feature constraint item, inputting the training set matrix into the model for solving and iterative updating, respectively training a dictionary and a coding coefficient matrix, and obtaining classifier parameters by using the coding coefficient matrix and a mark matrix of the training sample; and finally, for a test image, obtaining a sparse coefficient and a prediction label thereof through an orthogonal matching pursuit algorithm, and classifying various disease images by comparing the label of the reconstructed test set with the label of the original test set. According to the invention, classification of different lesion images of the gastrointestinal endoscope can be realized; endoscopic diseases can be effectively classified.

Description

technical field [0001] The invention relates to abnormal classification technology of gastrointestinal endoscope images, in particular to a dictionary learning method based on atomic class label embedding and local feature constraints of profiles, which is suitable for abnormal detection of gastrointestinal endoscope images. Background technique [0002] More and more people suffer from gastrointestinal diseases. As the gold standard for gastrointestinal research, endoscopy is widely used in early detection and adjuvant treatment of gastrointestinal diseases, effectively reducing morbidity and mortality. Conventional hand-held endoscopy is an invasive diagnostic device, and it is difficult to accurately obtain the situation of the entire gastrointestinal tract. Compared with traditional endoscopy techniques, wireless capsule endoscopy, as a new painless and non-invasive technology, not only can completely enter the small intestine, but also reduces the discomfort of patients...

AI Technical Summary

Problems solved by technology

[0004] Most wireless capsule endoscopy abnormality detection methods only consider a specific abnormality, and the existing multi-abnormality classification results are far from satisfactory. The dictionary learning method with standard embedding and profiles local feature constraints is used to classify the abnormalities of gastrointestinal endoscopy images. By combining the dictionary learning of multi-class images, the sparse representation of image features and the comparison of reconstruction errors, a multi-class dictionary is constructed. A linear classifier that can effectively classify different abnormal images of gastrointestinal endoscopy

Images