Extraction of characteristics from an image by analysis with multiple spatial resolutions

By analogy to the mammalian visual system, a scheme is presented for image characteristic extraction from several channels of analysis with spatial frequencies. The significant contours on several scales are integrated to form a connected sketch. Frequency bands are generated by the bias of a Gaussian filter to obtain a multiresolution pyramid. A top-down analysis is adopted to generate the final sketch. After choice of the vigilance index (number of layers of the pyramid used for analysis), information from a higher level is used to check the processing at the underlying level. The vigilance index controls the nature of information to be extracted. Integration of this contour detector into a system of automatic inspection of printed circuit boards shows that a small vigilance index makes it possible to verify the state of components, their identification and connection, whereas a large index leads to a sketch suitable for recognition of the components and verification of their correct positioning on the board