Recognition, object detection and segmentation of white background photos based on deep learning

Image recognition, object detection and segmentation have been a popular problem in computer vision tasks. This paper addresses the recognition, object detection and segmentation issues in white background photos with deep learning method. In particular, we firstly train a recognition model based on GoogLeNet to judge whether a photo is white background. Then we propose a main object detecting algorithm to eliminate unnecessary elements such as logos, characters with Faster R-CNN. Eventually a main object segmentation method combining both CRF-RNN network and Grabcut is adopted to smoothly eliminate the shadow area and obtain the fine segmentation results. All exploring algorithms are implemented in real time with Caffe and Tesla K80 from Nvidia. Latest testing experiments demonstrate that an accuracy of 96% in recognition and 94%in detection is resulted and an acceptable fine segmentation results used for solution of 400 ∗ 400 is successfully achieved. The proposed technical scheme offers an alternative to computer vision tasks in complex background photos.