Research On Multi-feature Medical Image Recognition Based On Data Fusion

Realizing the effective medical image recognition is a hot topic in the cross-research of computer science and medical image now. And the main factor of influencing the medical image recognition effect contains two aspects: feature extraction and recognition method.Medical image feature extraction is a foundational work for medical image recognition. On the basis of the in-depth analysis of medical image feature extraction algorithms home and abroad, this paper researched principally the Gabor wavelet texture feature extraction.After extracting kinds of features, how we use them to medical image recognition so that can get the satisfying recognition rate, this paper researched mainly on the multi-feature medical image recognition based on Data Fusion.The main contents in this paper can be summarized as the following three aspects:(1)This paper summed up and evaluated the methods of medical image color feature extraction, texture feature extraction, shape feature extraction and semantic feature extraction fully. Some typical feature extraction methods such as gray-scale histograms feature extraction, co-occurrence matrix feature extraction, invariant moment feature extraction and clustering feature extraction and so on, which were analyzed and researched in detail on MATLAB 7.0. (2)A new Gabor wavelet feature extraction algorithm based on effective data grid was researched and proposed. For the characteristic of Gabor wavelet feature extraction algorithm, and trying to make the extracted features express the medical image content better, the medical image grid-partition method was used, and a new algorithm of the Gabor wavelet texture feature extraction which was based on effective data grid was proposed, then the algorithm flow and experimental comparison were given.(3)A multi-feature medical image recognition algorithm based on data fusion was researched and proposed. As the shortcoming of the features of any class can not express the medical image efficiently, a new recognition framework combining the feature-level data fusion with the decision-level data fusion was proposed. As the results show, using PCA(Principle Component Analysis) method to carry out the feature-level data fusion, the fused features can express the medical image's content better, in the mean time, it can reduce the dimension and the redundancy of the features; and using voting method to carry out the decision-level data fusion, which could fully utilize the complementarity of different classifiers and a higher recognition rate can be got compared with single classifier.