Target enhancement using steerable pyramid decomposition and multi-channel filtering

Generally, targets are small as compared to the background and are obscured by the clutter. Therefore, for efficient automatic target recognition/detection (ATR/D) it is essential to enhance targets and reduce the background/clutter effect. In this paper, we describe a technique based on steerable pyramid decomposition and multi-channel filtering to enhance targets. The steerable pyramid decomposition offers the advantage of analyzing targets in different orientations and multi-resolution and hence, helps in extracting directional features across different bands that are mainly associated with targets. First, we preprocess the images (a) to remove the objects that are typically larger than the possible maximum size of the targets of interest and (b) to estimate the required levels of pyramidal decomposition. We then decompose the images into required levels by using a bank of steerable filters. Next, we apply a feature enhancement technique to all the sub-band images to generate enhanced feature based images. Finally, a novel multichannel filtering technique is applied to remove background clutter using these enhanced images. The result of filtering obtained at a lower resolution level is propagated into the higher resolutions. This has the advantage of greatly reducing the computation load and improving the target enhancement. We have applied the proposed algorithm to both simulated and real un-cooled infra red (IR) images. The experimental results show that the proposed algorithm is effective in removing the background clutter and enhancing the targets in both these cases.