Fast Detection of Distant, Infrared Targets in a Single Image Using Multiorder Directional Derivatives

The fast and robust detection of far targets is one of the key techniques in infrared searching and tracking applications. Using multiorder directional derivatives, an effective and concise detection method performed on a single IR image is proposed in this article. First, multiorder directional derivatives of an image are derived from the facet model. According to the derivative characteristics of small targets, enhancement filters are designed to maximize the target components. Convolved with these filters, the small targets are enhanced and the backgrounds are suppressed in each derivative subband. The final result is obtained by fusing all these filtered derivative subbands. The experimental results demonstrate that the proposed method consistently achieves a robust and effective performance on all datasets. In addition, the low computational complexity makes the proposed method suitable for real time detection purpose.

I. Introduction

Far target detection plays an important role in various applications, such as marine surface surveillance, air-to-air missile guidance, remote sensing in highly spatially structured optical imagery, infrared searching and tracking systems [1]–[6]. In infrared images, suspect airplane, missile or ship in long imaging distance usually presents as a small bright spot [7]. The timely detection of these targets contributes to accurate guidance, military defense, etc. However, these targets only occupy relatively small areas and their gray levels are usually dim, which means that the far targets lack effective shape, structure, and texture information [8]. In most references, these targets are also called small targets. Besides, backgrounds in infrared images are complex where much random noise and clutters exist [9], [10]. Therefore, it is challenging to detect small targets in such low signal-to-clutter ratio environments. Over the past few decades, extensive research efforts have been focused on this area and various algorithms have been proposed.