Object Segmentation via Semantic Mask and Color Clustering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for sky segmentation in images, either using deep learning algorithms or traditional color-based algorithms, suffer from partial missing segmentation and mis-segmentation, especially in images with small color differences.
Innovation Solution
An object segmentation method that involves obtaining an initial mask graph through semantic recognition, determining an initial target object area, performing clustering to obtain color classifications, generating difference graphs, and adjusting confidences to determine a target mask graph for accurate segmentation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If deep learning algorithm is used for sky segmentation, then segmentation capability is improved, but partial missing segmentation occurs in the middle of the segmented mask graph
Solution Approach 1:
The patent divides the segmentation process into two distinct stages: initial segmentation using deep learning to obtain a mask graph, and subsequent refinement using color space transformation and morphological operations to fill missing regions. This multi-stage segmentation approach resolves the contradiction by combining the strengths of different methods.
Solution Approach 2:
The patent performs preliminary segmentation using deep learning to obtain an initial mask graph, then uses this as input for subsequent refinement steps including color space transformation and morphological operations. This preliminary action allows the system to leverage the speed of deep learning while correcting its deficiencies in later stages.
2Productivity
If traditional algorithm based on color information is used for sky segmentation, then processing speed is improved, but mis-segmentation occurs relying on color of the sky
Solution Approach 1:
The patent merges two different segmentation approaches: deep learning-based initial segmentation and color-space-based refinement. By combining these methods, the system achieves both high processing speed from the deep learning model and high accuracy from the color-based refinement, resolving the contradiction between speed and precision.
Solution Approach 2:
The patent introduces color space transformation (LAB color space) and morphological operations as intermediary steps between the deep learning segmentation and final output. These intermediary operations refine the initial segmentation by leveraging color information without relying solely on it, thus improving accuracy while maintaining efficiency.
3Productivity
If traditional algorithm based on color information is used for sky segmentation, then processing speed is improved, but failure occurs with pictures with small color differences
Solution Approach 1:
The patent transforms the color information from RGB to LAB color space, changing the parameter representation of color. This transformation enhances the separation between sky and non-sky regions by emphasizing luminance (L channel) and color opponent dimensions (a and b channels), making the segmentation more robust to small color differences while maintaining processing efficiency.
Data Source
AI summary
Embodiments of the present disclosure disclose an object segmentation method and apparatus, a device and a storage medium. The method comprises: obtain an initial mask graph by performing semantic recognition on a target object in an image to be segmented; determining an initial target object area in the image to be segmented based on the initial mask graph; obtaining N color classifications of the target object by performing clustering processing on pixel points in the initial target object area according to color values; obtaining N difference graphs according to the N color classifications and the image to be segmented; determining a target mask graph according to the N difference graphs and the initial mask graph; and segmenting the target object in the image to be segmented based on the target mask graph.