Hybrid Histogram Descriptor: A Fusion Feature Representation for Image Retrieval

Area of Science:

• Computer Science
• Image Processing
• Artificial Intelligence

Background:

• Increasing availability of visual sensors generates vast image data, driving demand for efficient image retrieval.
• Developing effective feature representations is a critical challenge in image retrieval for diverse applications.

Purpose of the Study:

• To propose a novel fusion feature representation, the hybrid histogram descriptor (HHD), for enhanced image retrieval.
• To evaluate the performance and robustness of the HHD against existing methods.

Main Methods:

• The hybrid histogram descriptor (HHD) combines a perceptually uniform histogram (color and edge orientation) with a motif co-occurrence histogram.
• Performance was benchmarked on multiple datasets including RSSCN7, AID, Outex-00013, Outex-00014, and ETHZ-53.

Main Results:

• The HHD demonstrated superior effectiveness and robustness compared to ten recent fusion-based descriptors.
• The proposed descriptor achieved performance comparable to state-of-the-art convolutional neural network (CNN)-based descriptors.
• Computational complexity analysis confirmed the in-depth evaluation of the HHD.

Conclusions:

• The hybrid histogram descriptor (HHD) presents a powerful and efficient solution for content-based image retrieval.
• The HHD offers a competitive alternative to complex deep learning models, particularly when training data is limited.