Light Field Salient Object Detection Based on Spatial Feature Compensation

Abstract

The Light Field uses a new imaging technology that records both the intensity and direction of light in space. Compared to RGB and RGB-D images, light field data contains information about the color, intensity, position and direction of light, which is closer to the human perception of natural scenes. We propose a light field salient object detection algorithm based on spatial feature compensation(SFCNet). Firstly, this algorithm uses discrete gray-scale multi-view images as input to compress the amount of data. Based on this, we design a spatial feature aggregation module to integrate multi-view features and RGB features. In addition, a convolutional noise reduction module is designed to reduce the effect of noise in the depth information to address the problem of poor depth information. Finally, we discuss the way of multi-feature fusion for light field data. Experiments demonstrate that our algorithm achieves high accuracy detection results while effectively reducing the data size.