Enhanced “GrabCut” tool with blob analysis in segmentation of blooming flower images

Abstract

This paper discusses the enhancement using blob analysis applied to automatic segmentation of “GrabCut” tool [1] for segmenting blooming flowers in color images. The automatic segmentation of “GrabCut” is used to initialize the segmentation, but the results are not effective and there is insufficient separation of foreground and background color distributions. In our proposed work, the segmented “GrabCut” image in RGB format is first converted to a binary image based on the V plane of the HSV color space. The morphology operators combining with set operations are then applied to fill up the holes of blob. This is then followed by blob filtering to eliminate the unwanted connected region. Finally, the segmented binary image is converted back to its RGB form. The proposed enhanced method achieves a more efficient extraction of blooming flower in a complex environment which cannot be trivially eliminated by the automatic segmentation of “GrabCut”.