Detection of natural, variable objects

Detection of natural, variable objects requires general features that can be extracted from any image. Human vision is capable of segmenting images and recognizing objects in radiographs despite the summation of all objects in the inspection volume in the image plane. Thus, the features important in human perception provide a guide for automatic image processing. The initial features selected for evaluation are edges, defined by the wavelet transform and characterized by the Lipschitz exponent, and the fractal dimension, estimated by an efficient multiscale filter. Segmentation of the multiple parameter image, including spectroscopic absorption data, will then be effected using the Markov random field model and hierarchical agglomerative clustering algorithm of Panjwani and Healey. Military applications include internal inspection of munitions for manufacturing quality control, or possibly for treaty verification. Many security tasks require internal inspection for only generally defined objects, such as inspection for explosives of contraband. Commercial applications include inspection for ill-defined, high-resolution defects such as cracks and corrosion in sealed containers and plumbing. These image features can be used in the automated analysis of other types images.