A Superpixel-Level Semantic Annotation Method for Indoor Scenes

Abstract

A superpixel-level indoor scene semantic labeling method is disclosed, which can avoid the problem of huge computational cost of applying a deep network to pixel-level indoor scene labeling, and enables the deep network to accept a superpixel set as an input. This superpixel-level indoor scene semantic annotation method includes the following steps: (1) using a simple linear iterative clustering segmentation algorithm to perform superpixel segmentation on the indoor scene color image; (2) combining the indoor scene depth image with step (1) Obtained superpixels, extract superpixel kernel descriptor features (primary features); (3) construct superpixel neighborhoods; (4) construct superpixel depth network SuperPixelNet, learn superpixel multimodal features; treat labeled superpixels, Combined with the multi-modal features of the superpixel and its neighboring superpixels, superpixel-level semantic annotations are given for RGB-D images of indoor scenes.

Description

technical field [0001] The invention relates to the technical fields of multimedia technology and computer graphics, in particular to a superpixel-level indoor scene semantic labeling method. Background technique [0002] As a necessary work in computer vision research, semantic annotation of indoor scenes has always been a research hotspot and difficulty in the field of image processing. Compared with outdoor scenes, indoor scenes have the following characteristics: 1. There are many types of objects; 2. The occlusion between objects is more serious; 3. The scene is very different; 4. The illumination is uneven; 5. The lack of discriminative features . Therefore, compared with outdoor scenes, indoor scenes are more difficult to label. Indoor scene semantic annotation is the core content of indoor scene understanding. It has a wide range of applications in service, fire protection and other fields, such as robot mobile positioning and environment interaction, and event det...

AI Technical Summary

Problems solved by technology

[0009] In order to overcome the defects of the prior art, the technical problem to be solved by the present invention is to provide a super-pixel-level indoor scene semantic labeling method, which can avoid the problem of huge computational cost of deep network applied to pixel-level indoor scene labeling, and can Make a deep network accept a collection of superpixels as input

Images