Image Analysis Based Abnormality Detection Algorithm And Its Applications

Abnormality detection has been an important application in the area ofimage processing and pattern recognition. The methods of applying imageprocessing and machine learning algorithms for abnormality detection can notonly decrease manual workload and the differences between differentoperators, but also show higher sensitivity, lower false detection rate andhigher speed. The application has shown promising outlook in the areas suchas medical image processing.In this paper, after reviewing the general abnormality detection methods,the author mainly carries out throughout researches in two specific areas:bone scintigraphy images and X-ray security inspection systems images.To develop computer aided diagnosis system for bone scintigraphyimages, an adaptive algorithm for hotspot detection from spine and rib areasis first proposed. Other methods based on base-value and symmetry areapplied for whole body hotspot detection. After hotspot detection, theclassification problem for hotspots and patients is discussed. By featureextraction and classification, the probability for a hotspot being true positiveand a patient being metastasis can be evaluated. As shown in the experiments,the proposed algorithm shows a sensitivity of93.1%and the false positivedetections can be decreased as well. The computer aided diagnosis system based on the hotspot detection and classification algorithms has beendeveloped and is now under test, which can reduce the workload of thephysicians and intra-physician difference.On the other hand, new algorithms for material classification andcontraband detection are proposed for security inspection system based on thefusion of transmission and back scatter images. Firstly, the support vectormachine is used to classify all materials into three categories: organic matter,inorganic matter and mixture. Secondly, after image registration and fusion oftransmission and back scatter images, the SVM classifier is used for furthermaterial classification and contraband detection. The experiments show thatthe machine learning based method can greatly improve the accuracy formaterial classification and contraband detection.