Foreground Target Extraction in Bounding Box Based on Sub-block Region Growing and Grab Cut

• Main Authors: Yu, YaWei, Zhu, JiHong, Pei, JiHong
• Format: Conference Proceeding
• Language: English
• Subjects: Bounding box; Color clustering; Data mining; Deep learning; Feature extraction; Foreground extraction; Grab Cut algorithm; Image color analysis; Image segmentation; Machine learning algorithms; Object detection; Sub-block region growing
• ISSN: 2164-5221
• DOI: 10.1109/ICSP.2018.8652318

The region of interest for the network detection based on deep learning algorithms exists a problem of background interference, which goes against the feature extraction of the region of interest. In order to solve the problem of background interference faced by the region of interest, this paper proposes a foreground extraction algorithm for the target detection based on deep learning algorithms in combination of the sub-block region growing algorithm and the Grab Cut algorithm. Firstly, this algorithm pre-processes the color image through the Retinex algorithm to improve the image contrast; secondly, it extracts the features of the image sub-blocks (including dominant hue, secondary hue, color and luminance, etc.) and constructs the similarity measurement function between different sub-blocks; thirdly, it defines the seed region, sub-block region growing and termination criterion of the region growing and achieves the coarse segmentation of the foreground; finally, it re-extracts the coarse segmentation results that have been processed through the open operation by means of Grab Cut algorithm to obtain the foreground segmentation results of the input image. The experimental results show that this method can offer a better foreground extraction result with a low miss rate and a low false positive rate to solve the problem of background interference faced by the target detection based on deep learning network.