Continuous Wavelet Decomposition in Real-Time for Multi-Resolution Analysis

Wavelet representations of images has proven valuable for Multi-Resolution Analysis, but application of this technique has been limited by the computational requirements of the wavelet transform. The continuous wavelet transform can be formulated as a convolution, with a typical choice for kernel being the second derivative of the Gaussian. Convolving an image with this kernel is the intrinsic property of an analog circuit that can be implemented in a two-dimensional grid. Stimulating this circuit with a two-dimensional signal representing the pixel values of the image will produce an output corresponding to a wavelet decomposition of the image, all in real-time. By varying circuit control signals one achieves a continuous wavelet decomposition for arbitrary scale and shift parameters. Tanner Research proposes to develop an analog VLSI integrated circuit, the Real-Time Wavelet Transformer, for use in Multi-Resolution Analysis. It will accept input images and produce wavelet decompositions at video frame rates. Our revolutionary circuit concepts are efficiently implemented in analog VLSI, producing a Rcompact, high performance, low power signal processing component for desk-top and embedded image analysis applications. In Phase I we will design and fabricate an integrated circuit that will produce wavelet decompositions in real-time.