Image Segmentation Application of Conditional Convolution Restricted Boltzmann Machine

The traditional image segmentation method is not effective in characterizing the specific features of the target and the representation of high-level information, while the deep learning model doing this job well. In the existing Conditional Restricted Boltzmann Machine (CRBM), Restricted Boltzmann Machine (RBM) is used to capture the correlation of the output space. The condition represents the mapping relationship between features and categories. The deeper Conditional Deep Boltzmann Machine (CDBM) model developed on the basis of CRBM can obtain higher order information to improve the accuracy of image segmentation. However, the different image sizes in real life and various interference factors of images limit the performance of the above two types of models, and the convolution operation is outstanding in extracting the detailed local features of the image. Combining these two points, the Conditional Convolution Restricted Boltzmann machine (CCRBM) is proposed, further improving the accuracy and flexibility of image segmentation in shadow, clutter background and other situations of the images. The proposed method is tested in two datasets, the experiments results demonstrate the effective of the proposed method.