In this paper, an efficient approach for automatic and accurate sky region detection from fisheye images is proposed. The proposed approach starts by segmenting the acquired image into regions using Statistical Region Merging method. After that, the segmented regions are characterized using local RGB color descriptor using image quantization. The next step consists of classifying the characterized regions into sky and nonsky regions by using maximal similarity based region classification through Hellinger kernel based distance. In order to improve the obtained region classification results, a segment analysis based technique using Line Segment Detector is proposed. Experimental results prove the robustness and performance of the proposed procedure.