Wavelet-Transform Based Fast Underwater Salient Object Detection Algorithm

Abstract

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.