基于小波变换的水下显著性目标快速检测算法  

Wavelet-Transform Based Fast Underwater Salient Object Detection Algorithm

作　　者：孙欣悦 李庆忠[1] SUN Xinyue;LI Qingzhong(College of Engineering,Ocean University of China,Qingdao 266100,Shandong,China)

机构地区：[1]中国海洋大学工程学院,山东青岛266100

出　　处：《计算机工程》2023年第4期249-255,共7页Computer Engineering

基　　金：国家重点研发计划(2017YFC1405202);海洋公益性行业科研专项(201605002)。

摘　　要：目前地面显著性目标检测取得了较大进展,而水下场景具有较高的复杂性,导致水下显著性目标检测仍然面临诸多挑战。为了实现复杂水下环境的显著性目标快速检测,提出一种基于小波变换的水下显著性目标检测算法。对水下采集图像进行多级小波变换预处理,针对提取的低频子带图像,利用自适应中值滤波去除其中的斑点颗粒,对相应的高频子带进行显著性边缘检测以强化目标边缘信息。在此基础上,利用小尺度超像素分割与合并策略分割处理后的低频子带图像,通过基于区域对比度的显著性检测方法进行图像显著性计算。融合低频子带显著图和高频子带显著边缘图,得到最终的显著性检测结果。USOD公开数据集上的实验结果表明,在进行水下显著性目标检测时该算法的整体度量值达到93.9%,平均绝对误差低至3.08%,能较好地实现水下大目标和成群小目标的准确检测,且在处理大分辨率水下图像时具有良好的实时性,在CPU平台上每帧的显著性目标检测时间为168 ms,算法适用于水下机器人显著性目标快速检测应用场景。Ground-based salient object detection has made significant advances;however,the complexity of underwater scenes has led to many challenges in underwater salient object detection.For rapid detection of salient objects in complex underwater environments,an underwater salient object detection algorithm based on wavelet transform has been proposed in this study.The underwater image is preprocessed by multilevel wavelet transforms.For the extracted low-frequency sub-band image,the speckle particles are removed by an adaptive median filter,and the corresponding high-frequency sub-band is identified by significant edge detection to strengthen the object edge information.Based on this,the low-frequency sub-band image is segmented using a small-scale super-pixel segmentation and merging strategy,and the regional contrast-based saliency detection method is applied to calculate the image saliency.The final saliency detection result is obtained by fusing the low-frequency sub-band saliency and high-frequency sub-band saliency edge images.Experimentally,the Underwater Salient Object Detection(USOD)public dataset revealed that the overall measurement value of the algorithm is 93.9%with a Mean Absolute Error(MAE)of 3.08%,making it suitable for accurate detection of large underwater objects and small groups of objects while also providing good real-time performance in processing high-resolution underwater images.The average salient object detection time of each frame on the CPU platform was 168 ms,and the algorithm is applicable to the rapid detection of salient objects of underwater vehicles.

关 键 词：水下目标 显著性检测 小波变换 超像素分割 区域对比度 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]