Cosmic ray (CR) hits can affect a significant number of pixels both on long-exposure ground-based CCD observations and on the Space Telescope frames. Thus, methods of identifying the damaged pixels are an important part of the data preprocessing for practically any application. The paper presents an implementation of a CR hit detection algorithm based on a homogenous structure (also called cellular automata ), a concept originating in artificial intelligence and dicrete mathematics. Each pixel of the image is represented by a small automaton, which interacts with its neighbors and assumes a distinct state if it ``decides'' that a CR hit is present. On test data, the algorithm has shown a high detection rate (~0.7 ) and a low false alarm rate (<=0.1 ) in relatively little time per frame. A homogenous structure is extremely trainable, which can be very important for processing large batches of data obtained under similar conditions. Training and optimizing issues are discussed, as well as possible other applications of this concept to image processing.
Astronomical Data Analysis Software and Systems III
- Pub Date: