融合深度信息的室内RGB图像视觉显著物体快速检测方法  

An Indoor Object Fast Detection Method Based on Visual Attention Mechanism of Fusion Depth Information in RGB Image

作　　者：林昌[1] 何炳蔚[1] 董升升 

机构地区：[1]福州大学机械工程及自动化学院,福建福州350108

出　　处：《中国激光》2014年第11期205-210,共6页Chinese Journal of Lasers

基　　金：国家自然科学基金(51175087);福建省杰出青年基金(2013J06013);福建省海外留学基金

摘　　要：针对传统视觉注意机制在室内三原色(RGB)图像视觉显著物体检测中存在的运算复杂、检测精度低等缺点,提出了一种融合深度信息的室内RGB图像视觉显著物体快速检测方法。对室内RGB图像进行降采样和金字塔量化处理,从而降低图片的空间分辨率和计算复杂度。利用亮度、红绿以及黄蓝三通道的多特征视觉注意机制显著性检测模型以获得室内RGB图像的显著图。在显著图分析中提出显著区域生长策略,从而获得视觉显著区域的精确轮廓。融合深度信息获取视觉显著区域内显著物体数目以及显著物体相互之间的位置关系。通过室内场景实验,验证了方法的可行性和有效性。The traditional visual attention mechanism is complex and rough-detection for visual saliency detection indoor red-green-blue（RGB）image.In order to overcome these defects,a new fast visual saliency object detection method based on fusion depth information on indoor RGB image is proposed.A certain scale image is obtained by subsampling and pyramid-quantization to reduce the spatial resolution of the images so as to decrease the computational complexity.The intensity,red-green and yellow-blue three-channel features visual attention mechanism significant detection model is proposed to acquire saliency map.The saliency growing strategy is proposed to acquire the precise saliency region in the saliency analysis.The fusion depth information is utilized to detect the objects in salient region.The feasibility and effectiveness of the algorithm is verified in indoor detection experiments.

关 键 词：机器视觉 注意机制 显著区域 检测 

分 类 号：O436[机械工程—光学工程]