| 基于显著性目标检测的图像前景提取算法研究 |
 点此下载全文 |
| 引用本文：贾小云1，杨振英1，邵帆2，罗豪才1.基于显著性目标检测的图像前景提取算法研究[J].计算技术与自动化,2025,(4):110-115 |
| 摘要点击次数: 10 |
| 全文下载次数: 1 |
|
|
| 中文摘要:为解决自然场景中深度学习抠像算法依赖于人机交互且前景提取效果不佳的问题，提出了通过检测显著性目标的方式实现图像前景提取算法。该算法由显著性目标检测和alpha估计两个网络组成，前者在基准网络的基础上引入混合损失函数和深监督策略提高图像前景提取的完整度，后者通过增加上下文注意力引导模块的方式优化基准网络，实现对前景图像细节的精确恢复。改进算法以自然场景中人像作为待提取前景进行研究，在自制数据集上训练，公开数据集Composition-1K上测试。改进后算法在均方误差（MSE）指标上低至2.329×10-4平方像素，梯度误差指标上相较基准算法减小近60%。实验表明，改进后算法提升了头发、玻璃等半透明区域的alpha值估计的准确度从而提高图像中抠像精度，并且具有较强的鲁棒性与泛化能力，可应用于图像编辑等应用场景。 |
| 中文关键词:图像前景提取 显著性目标检测 alpha估计 混合损失函数 深监督 上下文引导 |
| |
| Research on Foreground Image Extraction Based on Saliency Object Detection |
|
|
| Abstract:To solve the problem of deep learning image matting algorithms in natural scenes relying on manual interaction and poor foreground extraction performance, a method of detecting salient targets is proposed to implement image foreground extraction algorithms. This algorithm consists of two networks: salient object detection and alpha estimation. The former introduces a mixed loss function and deep supervision strategy on the basis of the baseline network to improve the completeness of image foreground extraction, while the latter optimizes the baseline network by adding a context attention guidance module to achieve accurate restoration of foreground image details. The improved algorithm is studied using human portraits in natural scenes as the foreground to be extracted, trained on a self-made dataset, and tested on the publicly available dataset Composition-1K. The improved algorithm has the mean square error (MSE) index as low as 2.329 × 10-4 square pixels, and the gradient error index reduced by nearly 60% compared to the benchmark algorithm. The experiment shows that the improved algorithm improves the accuracy of alpha value estimation in semi transparent areas such as hair and glass, thereby improving the accuracy of image matting. It also has strong robustness and generalization ability, and can be applied in application scenarios such as image editing. |
| keywords:foreground image extraction saliency object detection alpha estimates mixed loss function deep supervision context guided |
| 查看全文 查看/发表评论 下载pdf阅读器 |
