Segmented-Region Based Approach Using Object Components Database to Detect Sky Region

In this paper, we propose a new approach to sky region detection. It enables robust detection for cloudy sky regions by evaluating similarity of visual features between combinations of segmented regions in an input image and sky region components stored in a database. To reduce the matching cost for the detection, the proposed approach adopts a combination search. It eliminates non-sky regions from the segmented regions by matching with sky region components in the database, and then restricts the number of segmented regions to be combined to that of the sky region components of the corresponding sky region in the database. A false detection-false rejection curve shows that 18% of the equal error rate is achieved for 300 nature images. The maximum reduction rate of matching cost is 1/50 using a database with 369 sky regions.