融合似物性前景对象与背景先验的图像显著性检测

计算机工程与科学

融合似物性前景对象与背景先验的图像显著性检测

郭鹏飞，金秋，刘万军

（辽宁工程技术大学软件学院,辽宁葫芦岛 125105）

收稿日期:2017-05-24 修回日期:2017-08-15 出版日期:2018-09-25 发布日期:2018-09-25
基金资助:
国家自然科学基金(61172144)；辽宁省教育厅科学技术研究一般项目(L2015216)

Saliency detection via objectness foreground

object and background prior

GUO Pengfei,JIN Qiu,LIU Wanjun

（School of Software,Liaoning Technical University,Huludao 125105,China）

Received:2017-05-24 Revised:2017-08-15 Online:2018-09-25 Published:2018-09-25

摘要/Abstract

摘要：

为了在复杂背景图像中准确地提取出图像的显著区域，提出一种结合似物性前景对象与背景先验知识的图像显著性检测方法（OFOBP）。该方法首先对图像进行超像素分割，计算超像素颜色空间分布，得到初始显著图；利用似物性检测方法获取多个目标窗口，由窗口建立搜索区域，结合二值化的初始显著图优化目标窗口；再利用多窗口特征对超像素做前景对象预测，获取前景显著图；其次建立背景模板，计算稀疏重构误差获取背景先验图；最后融合两种显著图，得到最终显著检测结果。在公开数据集上与11种算法进行比较，本文算法能够较为准确地检测出显著区域，尤其是在复杂背景下对多个显著目标的检测，存在明显的优势。

关键词: 显著性检测, 似物性检测, 超像素颜色空间分布, 窗口优化, 多窗口特征, 背景先验

Abstract:

In order to extract the salient region accurately from the image with complex background, we propose a saliency detection method based on the objectness foreground object and background prior (OFOBP). Firstly, the image is segmented by superpixels. The superpixel color spatial distribution of the image is calculated, and the initial saliency map is obtained. A certain number of target windows with corresponding target scores are obtained by the method of binarized normed gradients algorithm, and at the same time, the target windows are used to establish search areas. The multiwindow features are used to make foreground object forecast for superpixels so that the foreground saliency map is obtained. Secondly, the background template is established, and the background prior map is obtained by using the sparse reconstruction error. Finally, the two saliency maps are fused to get the final detection result. The effectiveness of the proposed method is verified by comparing it with other eleven algorithms in public data sets. The proposed algorithm can detect the salient regions more accurately, especially when dealing with the multiple salient object images with complex background.

Key words: saliency detection, objectness detection, superpixel color space distribution, window optimization, multiwindow feature, background prior

郭鹏飞，金秋，刘万军. 融合似物性前景对象与背景先验的图像显著性检测[J]. 计算机工程与科学.

GUO Pengfei,JIN Qiu,LIU Wanjun.

Saliency detection via objectness foreground

object and background prior

[J]. Computer Engineering & Science.

[1]	赵应丁, 岳星宇, 杨文姬, 张吉昊, 杨红云, . 基于多特征融合卷积神经网络的显著性检测[J]. 计算机工程与科学, 2021, 43(04): 729-737.
[2]	李伟生，陈曦. 一种结合显著性检测与词袋模型的目标识别方法[J]. 计算机工程与科学, 2017, 39(09): 1706-1713.
[3]	刘志伟，周东傲，林嘉宇. 基于图像显著性检测的图像分割[J]. J4, 2016, 38(01): 144-147.