Object-Based Image Retrieval Using the U-Net-Based Neural Network

Area of Science:

• Computer Science
• Artificial Intelligence
• Image Processing

Background:

• Digital image retrieval is increasingly important due to widespread internet and social media usage.
• Existing methods for content-based image retrieval (CBIR) face challenges in accuracy and efficiency.
• Deep learning techniques offer potential for improved feature extraction in image retrieval.

Purpose of the Study:

• To propose a novel U-Net-based neural network for image segmentation in CBIR.
• To integrate Haar Discrete Wavelet Transform (DWT) and lifting wavelet schemes for effective feature extraction.
• To enhance the accuracy and efficiency of content-based image retrieval systems.

Main Methods:

• A U-Net-based convolutional neural network (CNN) was employed for image segmentation.
• Haar DWT and lifting wavelet schemes were utilized for feature extraction.
• The proposed method was evaluated on two benchmark datasets: Corel 1K and Corel 5K.

Main Results:

• The proposed method achieved high accuracy rates of 93.01% on Corel 1K and 88.39% on Corel 5K.
• U-Net segmentation reduced feature vector dimensions and decreased feature extraction time by 5 seconds.
• Performance analysis demonstrated improvements in accuracy, precision, and recall for image retrieval.

Conclusions:

• The U-Net-based approach significantly enhances image retrieval performance.
• The integration of U-Net with Haar DWT and lifting wavelets offers a robust solution for CBIR.
• This research contributes to more accurate and efficient digital image retrieval systems.