Infrared Small Target Detection Based on Multi-scale Block Discrete Cosine Transform

Due to the variable target size and complex detection environment, small target detection for infrared images may be challenging for individual target analysis, making it easier to overlook alarm or false alarm issues. This study proposes an infrared small target detection technique based on multi-scale chunking discrete cosine transform, combining different characteristics of infrared images in the spatial and frequency domains. First, to improve the correlation between pixels, the multi-scale image-chunking model is applied in the spatial domain to chunk the image at various scales. Next, the chunked image goes through varying degrees of chunked discrete cosine transform reduction to eliminate background information by filtering the low frequency coefficients. Then the small target images is reconstructed to obtain after different scales of frequency filtering. Lastly, multi-scale fusion is carried out to produce the final small target detection outcomes.