Multidimensional Digital Filters for Point-Target Detection in Cluttered Infrared Scenes

Hugh L. Kennedy

DOI: https://doi.org/10.1117/1.JEI.23.6.063019

2015-01-17

Abstract:A 3-D spatiotemporal prediction-error filter (PEF), is used to enhance foreground/background contrast in (real and simulated) sensor image sequences. Relative velocity is utilized to extract point-targets that would otherwise be indistinguishable on spatial frequency alone. An optical-flow field is generated using local estimates of the 3-D autocorrelation function via the application of the fast Fourier transform (FFT) and inverse FFT. Velocity estimates are then used to tune in a background-whitening PEF that is matched to the motion and texture of the local background. Finite-impulse-response (FIR) filters are designed and implemented in the frequency domain. An analytical expression for the frequency response of velocity-tuned FIR filters, of odd or even dimension, with an arbitrary delay in each dimension, is derived.

Computer Vision and Pattern Recognition

What problem does this paper attempt to address?

This paper attempts to address the problem of detecting point targets in cluttered infrared scenes. Specifically, the paper focuses on how to automatically detect aerial targets at long distances using infrared sensors in complex backgrounds such as clouds, sea, terrain, or vegetation. Despite advances in thermal imaging and data processing technologies, background clutter still leads to high false alarm rates and low detection probabilities. The paper proposes a three-dimensional spatiotemporal prediction error filter (PEF) that enhances the contrast between the foreground and background by utilizing relative velocity, thereby improving target detection performance.

The main issues include:
1. **Background Clutter**: Traditional methods such as simple threshold detection or one-dimensional high-pass filters tend to produce high false alarm rates in dynamic backgrounds.
2. **Static Target Detection**: For static targets, traditional methods have a low detection probability.
3. **Spatiotemporal Consistency**: While two-dimensional spatial filters can effectively remove the background, they lose important temporal consistency and difference cues present in the biological visual system.
4. **Non-stationarity**: The edges and boundaries of targets in typical scenes lead to highly non-stationary signals, making traditional three-dimensional filters less effective.

The proposed method aims to find a balance between simple two-dimensional spatial and one-dimensional temporal high-pass filters and complex optimal three-dimensional filters. By using complex sine functions and freely choosing the support range and model order for each dimension, an effective three-dimensional prediction error filter is designed.