Fast image search with locality-sensitive hashing and homogeneous kernels map

Area of Science:

• Computer Science
• Machine Learning
• Image Processing

Background:

• Existing kernel locality-sensitive hashing (LSH) methods enable fast image searches but often compromise accuracy.
• Current LSH algorithms struggle to balance query speed with search result precision.
• There is a need for improved LSH techniques that maintain accuracy for efficient image retrieval.

Purpose of the Study:

• To enhance the accuracy of kernel locality-sensitive hashing (LSH) for image search.
• To develop a method that combines explicit feature maps with LSH for improved performance.
• To accelerate object classification and content-based image retrieval tasks.

Main Methods:

• Proposed a novel approach by applying explicit feature maps to homogeneous kernels.
• Integrated these feature maps with kernel locality-sensitive hashing (LSH).
• Validated the method on multiple large-scale datasets.

Main Results:

• The proposed method significantly improves search accuracy compared to existing techniques.
• Demonstrated enhanced performance in object classification tasks.
• Achieved faster and more accurate results in content-based image retrieval.

Conclusions:

• The integration of explicit feature maps with kernel LSH offers a superior solution for image search.
• This approach effectively addresses the accuracy limitations of previous LSH algorithms.
• The method provides a computationally efficient and accurate solution for large-scale image analysis.