CT And MRI Brain Image Fusion Algorithms Based On Sparse Representation

Medical images of different modes contain different information due to different imaging principles. The fusion of those images with complementary information can assist doctor in clinical diagnosis and treatment. Sparse representation theory develops very fast since21th century and this theory has already got great research achievements in the field of medical image fusion.This thesis studies the feature extraction and fusion of brain CT and MRI images based on sparse representation, and focuses on the image feature extraction based on learning dictionaries and activity measurement criteria in sparse domain, the main content includes:(1) Analyzing the research background, research significance, research status at home and aboard, and summarizing the image sparse representation theory.(2) Proposing an image fusion algorithm for brain CT and MRI images based on K-SVD and pulse-coupled neural network (Pulse-coupled Neural Network, PCNN).Medical image fusion algorithms based on spatial or transform domain have relatively low computational complexity, but pixel discontinuity, contrast reduction and edge blur are introduced in the fusion images, in order to solve this problem, an image fusion algorithm for brain CT and MRI images based on K-SVD and PCNN is proposed. The algorithm gets training samples by preprocessing technology, the learning dictionary and sparse coding of training samples are got by K-SVD algorithm and OMP algorithm; the l1-norm of sparse coding is input to motivate the PCNN, the output of this network is used to measure the activity level. The learning dictionary got by K-SVD algorithm can match the training samples well and the sparse coding based on this dictionary can present the feature of source images efficiently, what’s more, the activity measurement criteria based on PCNN conforms to the working principle of human’s visual system, so the performance of this algorithm is great.(3) Proposing an image fusion algorithm for brain CT and MRI images based on online dictionary learning (ODL).The K-SVD algorithm is second-order iterative batch procedure, accessing the whole training samples at each process of dictionary update, when the number of training samples is very large, its running speed is very low, in order to solve this problem, an image fusion algorithm for brain CT and MRI images based on online dictionary learning is proposed. This algorithm gets the training set by preprocessing technology, the learning dictionary and sparse coding of training samples are got by online dictionary learning algorithm and LARS algorithm; the l1-norm of sparse coding is used to measure activity level, the sparse coding is fused by maximum activity criteria. The speed of this algorithm improves significantly because the online dictionary learning algorithm accesses a very small subset of the training samples.