A Light Field Salient Object Detection Method Based on Deep Convolutional Network

Abstract

The invention discloses a method for detecting salient objects in a light field based on a deep convolutional network. The steps include: 1. Converting sub-aperture images of all viewing angles from light field data obtained by using a light field acquisition device; 2. The sub-aperture image is reconstructed into a microlens image; 3. Data enhancement is performed on the microlens image; 4. Based on the pre-trained weights of the Deeplab‑V2 network, a salient target detection model combined with the microlens image is built, and the data set is used for training; 5. The trained salient object detection model performs salient object detection on the light field data to be processed. The method of the invention can effectively improve the accuracy of salient target detection in complex scene images.

Description

technical field [0001] The invention belongs to the fields of computer vision, image processing and analysis, and specifically relates to a method for detecting a prominent object in a light field based on a deep convolutional network. Background technique [0002] Salient object detection is a perceptual capability of the human visual system. When observing an image, the vision system can quickly acquire the regions and objects of interest in the image, and the process of obtaining the regions and objects of interest is salient object detection. With the development of computer technology and the Internet, as well as the popularization of mobile smart devices, people's acquisition of external images has shown a blowout growth. Salient object detection selects a small part from a large amount of input visual information to enter subsequent complex processing, such as object detection and recognition, image retrieval, image segmentation, etc., which effectively reduces the c...

AI Technical Summary

Problems solved by technology

[0008] 1. In the two-dimensional salient target detection method, since the two-dimensional image is the integral of light projected on the camera sensor, which only contains the light intensity in a specific direction, the two-dimensional salient target detection is too sensitive to high-frequency parts or noise. And it is easily affected by factors such as foreground and background color texture similarity, background clutter, etc.

[0009] 2. In the 3D salient target detection method, the accuracy of scene depth information depends on the depth camera. The current depth camera has low resolution, narrow measurement range, large noise, inability to measure transmissive materials, and is susceptible to interference from sunlight and smooth surface reflections. and many other issues

[0010] 3. In the 3D salient target detection method, feature information such as color, depth, and position are processed and fused independently of each other, and their complementarity is not considered comprehensively.

[0011] 4. Most salient object detection methods based on two-dimensional and three-dimensional images are based on the assumption that there is a clear difference between the object and the background, and the background is simple. With the large-scale increase of image data, the complexity of image content increases. These methods exist certain limitations

[0012] 5. In light field salient object detection, the research on light field data in salient object detection has just started, and currently available data sets are less and the image quality is poor

The current salient target detection using light field data is based on the traditional salient feature calculation method, and at the same time, multiple clues such as color, depth, and refocus are modeled separately, and there are problems such as insufficient feature expression and poor robust detection effect.

Images