Super-resolution image reconstruction method and device

Abstract

The invention provides a super-resolution image reconstruction device. The device comprises a training sample obtaining unit, a dictionary construction unit, a low-resolution image input unit, a sparse encoding unit, an image updating unit and a super-resolution image reconstruction unit, wherein the training sample obtaining unit is used for obtaining training samples; the dictionary constructionunit is used for constructing a dictionary; the low-resolution image input unit is used for converting low-resolution images into high-resolution images to carry out initial estimation; the sparse encoding unit is used for carrying out sparse encoding on each image patch in a detail layer of the currently estimated high-resolution image according to the dictionary; the image updating unit is usedfor updating the detail layer of the currently estimated high-resolution image; and the super-resolution image reconstruction unit is used for storing the updated high-resolution image to carry out estimation when iterative solution is in a convergence state, and otherwise, circularly and iteratively carrying out sparse encoding on each image patch in the detail layer in high-resolution image estimation. The invention furthermore discloses a corresponding super-resolution image reconstruction method. Through the super-resolution image reconstruction method and device, magnetic resonance imageresolution can be remarkably improved, distortions such as image noise and blurring can be effectively removed, complicated fine structures can be recovered and better subjective and objective effects can be provided.

Description

technical field [0001] The invention belongs to the field of image processing, and in particular relates to a super-resolution image reconstruction method and device. Background technique [0002] With the rapid development of magnetic resonance imaging technology, the resolution, signal-to-noise ratio and scanning speed of magnetic resonance images have been greatly improved. Unsatisfactory problems are still not well resolved. In neonatal magnetic resonance imaging, a single voxel may contain several different types of tissue, and external disturbances tend to cause restlessness in the neonate in the magnetic resonance imaging environment, these factors will further reduce the resolution of the captured neonatal magnetic resonance image . [0003] In order to improve the resolution of neonatal magnetic resonance images and the diagnostic efficiency of doctors, super-resolution reconstruction of neonatal images is generally performed by interpolation methods in clinical p...

AI Technical Summary

Problems solved by technology

But the inner image is from the input degraded low-resolution image, while the outer image is from a general large-scale image set, the global and local patterns in these inner and outer images may not be compatible with the target high resolution, resulting in the target High resolution image reconstruction error

Existing super-resolution methods based on deep learning require a large number of image sets and corresponding labels to train convolutional neural network models or generate adversarial network models, but in reality, newborn magnetic resonance images are not easy to obtain and the number is small, and they have not considered Longitudinal image prior knowledge modeling of the same individual or different individuals constrains the super-resolution performance of the trained model

Therefore, the anatomical accuracy and resolution of neonatal MRI images reconstructed by existing super-resolution methods are still not ideal.

Images