Accurate lesion segmentation based on endoscopy images is a fundamental task for the automated diagnosis of gastrointestinal tract (GI Tract) diseases. Previous studies usually use hand-crafted features for representing endoscopy images, while feature definition and lesion segmentation are treated as two standalone tasks. Due to the possible heterogeneity between features and segmentation models, these methods often result in sub-optimal performance. Several fully convolutional networks have been recently developed to jointly perform feature learning and model training for GI Tract disease diagnosis. However, they generally ignore local spatial details of endoscopy images, as down-sampling operations (e.g., pooling and convolutional striding) may result in irreversible loss of image spatial information. To this end, we propose a multi-scale context-guided deep network (MCNet) for end-to-end lesion segmentation of endoscopy images in GI Tract, where both global and local contexts are captured as guidance for model training. Specifically, one global subnetwork is designed to extract the global structure and high-level semantic context of each input image. Then we further design two cascaded local subnetworks based on output feature maps of the global subnetwork, aiming to capture both local appearance information and relatively high-level semantic information in a multi-scale manner. Those feature maps learned by three subnetworks are further fused for the subsequent task of lesion segmentation. We have evaluated the proposed MCNet on 1,310 endoscopy images from the public EndoVis-Ab and CVC-ClinicDB datasets for abnormal segmentation and polyp segmentation, respectively. Experimental results demonstrate that MCNet achieves 74% and 85% mean intersection over union (IoU) on two datasets, respectively, outperforming several state-of-the-art approaches in automated lesion segmentation with endoscopy images of GI Tract.

While the development of steganalysis algorithm keeps progressing, the traditional image steganagraphy has confronted with a security bottleneck, as this secret embedding destroys the naturality of cover image. To overcome this,the coverless image steganography(ClS) is proposed to select images containing desired secrets, and it has received widespread attention because it does not modfy the cover image to hide the secret information, and which keeps the nature of the image lisel.However, the existing CIS methods are restricted by the size of image database,and up to now,the maximum hidden capacity is only 18 bits. To break this limitation, this paper proposes a new ClS method based on image selection and StarGAN, in which, a traditional CIS method is used to select a natural image from database to represent the first part of the secret information, and then a mapping is established between the rest of the secret information and the face attrbutes.Finally, the StarGAN is used to generate a high-quality stego image with the mapping. Compared with the existing ClS methods, the proposed method increases the hidden capacity and maintains a better image quality with a strong robustness and security performance.