Research On Multi-feature Fusion Technology And Its Application In Medical Image Recognition

With the continuous development of Computer Vision and a variety of advanced medicalimaging equipment, the information contained in the medical image is very rich. It is of greatsignificance for clinical automation diagnosis. Single image feature is difficult to fully expressmedical image content.Multi-feature fusion has become a natural way to extract the medicalimage features. It can comprehensively utilize the medical image information to gain morerapid and accurate understanding of medical images.From low to high, information fusion can be divided into three levels. The feature-levelfusion not only keeps the most original information, but also overcomes the unstable and largecharacteristics of original data. Fusion feature can be effectively used in medical imagerecognition.Firstly, gray histogram features, color moment features, GLCM features, discrete wavelettransform features, and moment invariant features are researched in detail and implemented,which is the basis for the feature fusion. Secondly, principal component analysis (PCA)method based on multivariate statistical analysis is used in feature-level fusion. And it isapplied in liver B-image recognition. The recognition results are analyzed and compared.Finally, through the analysis and research of liver B-image feature extracted, there is obviouscorrelation in some dimension of the initial feature. The high-dimensional feature extracteddirectly applys to the fusion process, which will increase the time complexity in the laterprocessing.Fuzzy method to implement feature crude selection is proposed.Then the PCA isused for feature fusion.And it is applied in the liver B image recognition. Recognition effectsare compared from average accuracy and recognition time performance.The experimental results show that fusion feature can fully and effectively expressmedical image, which can bring better recognition results.Analyzing and comparing the feature selection results of different sample images, the results show that feature selection isstable and effective. Comparing with the results of direct PCA fusion applications, therecognition effect after feature selection is better, not only improve s the the average accuracyrate of recognition but also reduces the time complexity of the recognition process.It hasbetter performance, can be more effectively applicated in medical image recognition.Medicalimage recognition system is developed using Visual C++6.0and OpenCV1.0, which achievesliver B-image pre-processing, feature extraction, feature selection and fusion method. It has acertain significance and practical value in the computer-aided diagnosis.