Human visual system based processing for high resolution remote sensing image segmentation

Image segmentation is very essential and critical to image processing and pattern recognition. Watershed is the most popular one among all the proposed image segmentation algorithms, but it suffers from over-segmentation. To resolve the over-segmentation problem and obtain a concise region representation has been the focus of many researchers. There are many ways to reduce the over-segmentation. However, Human visual system (HVS) is often not incorporated into the consideration and process of segmentation. This paper presents a new approach to high resolution remote sensing image segmentation taking into consideration human visual system (HVS) model. A Contrast Sensitivity Function (CSF) based filtering is applied to the image before watershed transform. Multi-scale spatial frequency filtering images are derived from setting multi-scale viewing distance. Then the region merging based on Just Noticeable Difference (JND) is applied to the segmentation results. Finally, the effect of reducing over-segmentation based on CSF and JND is analyzed and significant improvement is reported in the experimental results.