An Improved Approach in Shadow Detection and Reconstruction INVHSR Images

Background/Objectives: The objective of this study is detection and reconstruction of shadows from Very High Spatial Resolution Images (VHSR images). Methods/Statistical Analysis: This paper, presents an effective unsupervised method of segmentation for detecting shadows using modified Self Organizing Maps (SOM and MRF). This term MRF is adapted with SOM to segment the shadows without giving more trained sample data set and linear regression method is applied in shadow regions for compensating the shadows by non shadow regions (Reconstruction process). Findings: Experimental result of proposed method gives a significant result on VHSR images and assess the various quality metrics to obtain the better performance, quality and accuracy than the other conventional methods. Application/Improvement: The Modified SOM network eliminates the fake shadows, avoids over segmentation reduces the execution time for segmenting the shadows and improves accuracy as appx 92%. It is mainly used for various remote sensing applications such as change identification, object recognition, scene restoration, color tuning etc.