DSP ALGORITHMS

Our work in the area of digital signal processing involves using DSP chips to perform pattern recognition algorithms. For example, the results for the vector inner product(VIP) operator approach are summarized as follows.

VIP OPERATOR RESULTS

Figure 1: Input Scene. The top two images are shuttle space ships (target class) and the bottom two are Hubble telescope images (non-target class). The yellow grid indicates relative translation offsets of the images and was not part of the input scene.

Figure 2: Output Response. Note that positions of 2 targets and 2 non-targets correspond to the input scene.

Figure 3: Output Response. 3-D representation of Figure 2.

Figure 4: Output target and non-target response versus number of VIP operators used in the operator bank. The top and yellow line represent the target and non-target peak response using a Matched filter. The top and red line represent the response as a function of the number of VIP operators used.

Figure 5: Numerical Performance. The yellow line represents the number of operations necessary to perform FFT based, Matched filtering. The green line represents the number of operations necessary to perform the VIP operations. Input and output SNRs are constant while the horizontal axis indicates the pixel width of an NxN input scene. For the FFT method, the operation order is N^2 log(N) while the VIP method is N^2.