Salient Object Detection by LTP Texture Characterization on Opposing Color Pairs under SLICO Superpixel Constraint

Area of Science:

• Computer Vision
• Image Processing
• Artificial Intelligence

Background:

• Salient object detection is crucial for computer vision applications.
• Current models struggle with color and textured images, often treating color and texture independently.
• A unified approach is needed to effectively process these features.

Purpose of the Study:

• To develop an efficient and robust salient object detection model.
• To integrate color information with local textural patterns for improved saliency mapping.
• To create a simple model with minimal parameters for practical applications.

Main Methods:

• Utilized Local Ternary Patterns (LTP) on opposing color pairs to characterize color micro-textures.
• Employed Simple Linear Iterative Clustering (SLICO) for superpixel segmentation.
• Applied FastMap, a variant of Multi-dimensional Scaling (MDS), to compute dissimilarity between color micro-textures.
• Combined saliency maps from multiple color spaces (RGB, HSL, LUV, CMY) for final output.

Main Results:

• The proposed model generates robust saliency maps by effectively integrating color and texture.
• Evaluations on five standard datasets using Mean Absolute Error (MAE), Mean Squared Error (MSE), and Fβ measures demonstrated superior performance compared to state-of-the-art models.
• The model achieved high accuracy in salient object detection.

Conclusions:

• The novel strategy of integrating color into local textural patterns offers an efficient solution for salient object detection.
• The simple yet powerful model outperforms existing methods.
• This approach can be combined with other techniques for further performance enhancement.