| 摘要: |
| 目的 近年来,随着数字技术的发展,图像篡改已经成为一个日益严重的问题,现有的众多图像篡改和目标检
测方法均存在 着 识 别 精 度 不 足 和 检 测 效 果 不 理 想 的 问 题。 为 了 更 有 效 解 决 这 一 挑 战, 提 出 一 个 基 于 改 进 的
YOLOv8 检测算法,希望能够实现更高的检测精度。 方法 首先,为了捕获篡改的边缘特征,引入局部和全局注意力
机制,对 YOLOv8 的骨干网络进行深度优化,这种优化结合了上下文感知和局部增强技术,大大增强了对边缘特征
的识别力;考虑到篡改区域可能存在各种形态,进一步采用堆叠特征金字塔的网络结构,以捕获多尺度特征;最后,
为了提高模型的计算效率和推理速度,在模型中结合深度可分离卷积和通道重排技术。 结果 在一系列的实验中,
改进的 YOLOv8 篡改检测算法在 CASIA2. 0 图像篡改数据集上展现了出色的性能,与原始算法相比,其准确率高达
82. 3%,明显提高了检测的效果;提出的基于改进的 YOLOv8 篡改检测算法,经过深度的网络优化和结构调整,成
功地提高了图像篡改检测的准确性和效率。 结论 文中提出的方法在图像篡改检测方面表现出较高的准确性,为图
像篡改检测领域带来了重要的进展,具有深远而重要的意义。 |
| 关键词: 图像篡改 目标检测 YOLOv8 下文感知 多尺度特征 深度可分离卷积 |
| DOI： |
| 分类号: |
| 基金项目: |
|
| An Improved YOLOv8 Algorithm for Image Tampering Detection |
|
XUAN Gaoyuan
|
|
School of Artificial Intelligence Anhui University of Science and Technology Anhui Huainan 232001 China
|
| Abstract: |
| Objective With the advancement of digital technology in recent years image tampering has emerged as a
growing concern. Many existing methods for image tampering and object detection suffer from inadequate recognition
accuracy and unsatisfactory detection effects. To address this challenge more effectively a detection algorithm based on an
improved YOLOv8 was proposed aiming to achieve higher detection accuracy. Methods Firstly to capture the edge
features of tampering a local and global attention mechanism was introduced to deeply optimize the backbone network of
YOLOv8. This optimization combined context awareness and local enhancement techniques significantly enhancing the
recognition of edge features. Considering the diverse shapes of tampered regions a network structure of stacked feature
pyramids was further adopted to capture multi-scale features. Finally to improve the computational efficiency and
inference speed of the model deep separable convolutions and channel shuffling were integrated into the model. Results
In a series of experiments the improved YOLOv8 tampering detection algorithm demonstrated excellent performance on the
CASIA2. 0 image tampering dataset. Compared with the original algorithm it achieved an accuracy as high as 82. 3%, significantly enhancing detection effectiveness. The proposed tampering detection algorithm based on the improved
YOLOv8 through deep network optimization and structural adjustments successfully improved both the accuracy and
efficiency of image tampering detection. Conclusion The proposed method in the paper demonstrates high accuracy in
image tampering detection representing significant advancements in this field. This study has profound and meaningful
implications for the field of image tampering detection. |
| Key words: image tampering target detection YOLOv8 context-awareness multi-scale features depth-separable
convolution |
