Guilin Yao

Image matting is a fundamental task in the field of computer vision, aiming to separate the foreground and background in an image. Existing studies have demonstrated that the matting method based on Plain ViT already has the ability to precisely segment images. However, the current methods have problems such as the decoder lacking feature selection and insufficient information integration when restoring resolution, which can lead to the loss of some details in the matting results. Based on the above issues, this paper proposes a matting network based on Plain ViT and hierarchical attention feature fusion mechanism. By redesigning the decoder, it completes the repair of matting details. During decoding, the decoder improves network performance by filtering feature channels, using dilated convolution, and applying attention mechanisms for low-level feature aggregation. Experimental results show that HFF-Matte has significantly improved accuracy compared to previous methods without significantly increasing the number of parameters.

Three-dimensional human pose estimation (3D HPE) aims to recover the three-dimensional coordinates of human joints from 2D images or videos to achieve precise quantification of human movement. In 3D HPE tasks based on multi-class complex human action datasets, the performance of existing Graph Convolutional Network (GCN) and Transformer fusion models is constrained by the fixed physical connections of the skeleton, which impedes the modeling of cross-joint long-range semantic dependencies and hinders further performance gains. To address this issue, this study proposes a semantic prior-based non-Euclidean topology enhancement method for multi-class complex human actions, built upon a GCN–Transformer fusion model. The proposed method retains the original physical connections while introducing semantic prior edges; by constructing a hybrid topology structure, it explicitly models long-range semantic dependencies between non-adjacent joints, thereby facilitating the extraction of cross-joint semantic information. Experimental results on the Human3.6M and HumanEva-I datasets surpass those of SOTA baseline models. On the Human3.6M dataset, MPJPE and P-MPJPE are reduced by 1.25% and 0.63%, respectively. For the Walk and Jog actions on the HumanEva-I dataset, MPJPE is reduced by approximately 6.5%. These results demonstrate that the proposed method offers significant advantages for 3D HPE tasks based on multi-class complex human action data.

Image segmentation, a fundamental problem in image processing, involves distinguishing the foreground from the background. Traditional image segmentation methods are typically divided into local and global approaches. Local methods often result in blurred segmentation due to their reliance on overly localized color sampling models. Conversely, global methods, which depend on the overall color distribution, struggle with images where the foreground and background share similar global characteristics. To overcome these limitations, a novel image segmentation method that integrates nonlocal criteria with graph cut theory is proposed. Initially, the method transforms the image’s color space to the HSV color space, enhancing color contrast. Following this, nonlocal criteria are utilized to compute the data term within the graph cut model, which is then incorporated into the traditional graph cut framework for final segmentation. This approach broadens the neighborhood sampling range in local models, allowing the model to identify more valuable nonlocal samples while mitigating the impact of color similarity in global models. Experimental results across various datasets indicate a marked improvement in segmentation quality achieved by the proposed method.