Linear models of cumulative distribution function for content based medical image retrieval

Area of Science:

• Computer Vision
• Image Processing
• Pattern Recognition

Background:

• Accurate and efficient image matching is crucial for various applications.
• Traditional methods often struggle with variations in image size and computational complexity.
• Cumulative Distribution Functions (CDFs) offer a robust statistical representation of image data.

Purpose of the Study:

• To develop a novel image matching technique based on Cumulative Distribution Functions (CDFs).
• To enhance the efficiency and scalability of image comparison algorithms.
• To enable robust matching of images irrespective of their size.

Main Methods:

• Approximating query and database image CDFs using piecewise linear models.
• Employing a two-parameter model (slope and intercept) for CDF approximation.
• Utilizing least squares line fitting for parameter estimation.
• Comparing images based on estimated CDF slopes and intercepts.

Main Results:

• The method allows comparison of images with different sizes due to the CDF's dynamic range (0 to 1).
• Approximation of CDFs with lines reduces feature dimensionality, improving matching speed.
• The piecewise linear CDF approximation facilitates a hierarchical matching methodology.
• The method demonstrates robustness and efficiency in image matching tasks.

Conclusions:

• The proposed CDF-based image matching method offers a computationally efficient and scalable solution.
• Piecewise linear approximation of CDFs provides a robust feature representation for image comparison.
• This technique is well-suited for large-scale image databases and real-time applications.