基于深度信息的室内显著物图像特征检测仿真

Simulation of Indoor Saliency Image Feature Detection Based on Depth Information

针对现有的室内显著物图像特征检测方法存在的检测准确率低、漏检率较高的问题,提出了一种基于深度信息的室内显著物图像特征检测方法,以室内显著物图像的法向量作为特征进行边缘检测,检测出边缘后将二维梯度作为主要的图像深度信息特征表量,通过分析特征融合与二维梯度直方图,完成对基本区域的分割。将分割后的图像作为室内显著物图像特征检测的样本,通过背景预测计算出图像中除显著物外的背景区域,利用深度卷积神经网络预测室内显著物图像的背景分布。采用对比度方法将局部邻域或者整个显著物图像作为对比区域,结合对比度法与显著性深度信息特性将背景区域作为对比度参考区域,在得到背景区域后,求解像素的显著值和权重值,实现对室内显著物图像特征检测。仿真结果表明,所提方法对显著物图像特征检测的效果很好,可以满足用户对显著性图像特征检测高准确率、低漏检率的要求。

Due to low detection accuracy and high miss detection rate, this article proposed a method to detect the indoor salient object image feature based on depth information. At first, we used the normal vector of indoor salient object image as the feature of edge detection. After detecting the edge, we used the two-dimensional gradient as the main quantity-presentation of image depth information feature. By analyzing the feature fusion and the two-dimensional gradient histogram, we completed the segmentation for basic region. Furthermore, we used the segmented image as the sample to detect the indoor salient object image feature and calculated the background region in addition to the salient object by background prediction. Moreover, we used deep convolution neural network to predict the background distribution of indoor salient object image. After that, we used the contrast method to take the local neighborhood or the whole salient object image as the contrast region. Combined the contrast method with the saliency depth information characteristic, we used the background area as the contrast reference area. After the background area was obtained, we found the significant value and the weight value of pixel. Thus, the feature detection of indoor salient object image was realized. Simulation results show that the proposed method has good effect on the feature detection of salient object image, which can meet the requirements of high accuracy and low miss detection rate for detecting the salient object image feature.