Abstract

The wavelet transform of an image may be viewed as a series of bandpass filtered versions of the image where each bandpass filter is a scaled version of a prototype filter. It provides a method of viewing the image at various spectral resolutions. The wavelet transform is being investigated as a method for inspecting textured industrial products produced in rolls or sheets (paper, metals, etc.) for flaws. An optical method to implement the 2-D wavelet transform in parallel has been developed. A coherent optical processor is used to generate multiple bandpass filtered images of an input image. The measured output is the magnitude squared of the wavelet images, or the "scalogram." An electronic computer can then be used to analyze the wavelet images. The goal is to locate and classify flaws in products by noting significant deviations in the statistics of the wavelet images from the expected image statistics.

© 1992 Optical Society of America