图像显著性区域检测模型研究及其应用

发布时间：2018-11-20 20:39

【摘要】：随着计算机与通信技术的快速发展,图像和视频日益成为承载数据信息的主要形式。图像资源出现爆炸式增长,如何通过计算机快速处理和检索图像是面临的重大挑战。显著性检测技术是图像检索、图像自适应分割、目标识别等计算机视觉领域的重要步骤。人类视觉系统善于帮助人们在面对复杂场景时搜索到自己感兴趣的区域,模拟人类的视觉机制提取与目标相关的显著区域,可以为显著区域优先地分配图像合成与分析所需的计算资源,提高计算机处理图像的效率。在研究已有的显著性检测模型的基础上,总结出现有模型在提取显著区域时面临着准确度不高,提取轮廓不清晰,抗噪能力差等缺点,所以研究精确的显著性检测模型变得很重要。本文主要研究工作如下:(1)结合局部特征对比突出显著物体边缘和全局特征对比突出内部区域的优点本文提出了一种应用局部特征和全局特征对比的显著性检测模型(LGC模型)。该算法首先使用简单的线性迭代聚类(Simple Linear Iterative Clustering)分割算法将图像预分割为若干紧凑的超像素,选取边界区域集并计算所有超像素的边界权重;然后计算颜色和纹理特征的局部对比度得到局部显著图,利用全局特征的独特性,空间分布特性得到全局显著图;最后采用求和乘积(Sum and Product)方法将局部和全局显著图融合得到最终的显著图。在Achanta测试集上进行对比分析,实验结果表明,与其它5种算法相比本文显著性检测算法准确度更高,具有较大的优势。(2)将本文提出的显著性检测模型,应用在图像感兴趣区域自动分割、内容敏感的图像缩放以及非真实性渲染应用中。实验结果表明本文提出的模型相对于传统的模型在图像处理应用中的效果更好。
[Abstract]:With the rapid development of computer and communication technology, image and video increasingly become the main form of carrying data information. With the explosive growth of image resources, how to quickly process and retrieve images by computer is a major challenge. Salience detection is an important step in the field of computer vision such as image retrieval, image adaptive segmentation and target recognition. Human visual systems are good at helping people find areas of interest to themselves in the face of complex scenes, and simulate human visual mechanisms to extract salient areas associated with targets. The computational resources needed for image synthesis and analysis can be allocated first for significant regions, and the efficiency of computer processing can be improved. On the basis of studying the existing salience detection models, it is concluded that the models are faced with some shortcomings, such as low accuracy, unclear contour, poor anti-noise ability and so on. So it is very important to study the accurate salience detection model. The main research work of this paper is as follows: (1) combining the advantages of local feature contrast prominent object edge and global feature contrast highlighting internal region this paper proposes a significant application of local feature and global feature contrast. Detection model (LGC model). Firstly, a simple linear iterative clustering (Simple Linear Iterative Clustering) algorithm is used to predivide the image into several compact super-pixels, then the boundary regions are selected and the boundary weights of all super-pixels are calculated. Then the local contrast of color and texture features is calculated and the global salience map is obtained by using the uniqueness of the global feature and the spatial distribution characteristic. Finally, the sum product (Sum and Product) method is used to fuse the local and global salient graphs to obtain the final significant graphs. The experimental results on the Achanta test set show that compared with the other five algorithms, the significance detection algorithm in this paper has higher accuracy and more advantages. (2) the significance detection model proposed in this paper is proposed. It is used in automatic segmentation of region of interest, content sensitive image scaling and non-realistic rendering applications. The experimental results show that the proposed model is more effective than the traditional model in image processing.
【学位授予单位】：长沙理工大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41

【参考文献】