Content Based Image Retrieval

Inputs:

• Target image (T)
• Image database (B)
• Method for capturing feature (Feature type) (F)
• Distance metrics to compare feature from two images(D(Ft,Fi))
  • Sum of square distance
  • Histogram intersection
• Desired number of output images N

Process:

• Compute the features Ft on the target image T
• Compute the features Fi on all the images in B
  • Executing many queries with different target images
  • Features are computed and stored in a CSV file and then used for many different target image
• Compute the distance of T from all the images in B using the distance metric D(Ft,Fi)
• Sort the images in B according to their distance from T and return the best N matches

Methods to find similar images:

• Baseline Matching
  • 7x7 square in the middle of the image as feature vector
  • Sum of square differences as the distance metric.
• Histogram Matching
  • Single normalized color histogram. (2D) as feature vector.
    • Whole image r,g chromaticity histogram using 16 bins for each r and g
    • Whole image r,g chromaticity histogram using 8 bins for each r and g
  • Histogram intersection as distance metric.
• Multi-Histogram Matching
  • 2 color histogram as feature vector with 8 bins
  • The histogram represents different spatial parts of the image.
    • Overlapping
    • Disjoint
  • Histogram intersection as distance metrics and weighted averaging to combine the distances between the different histograms
• Texture and color
• Deep Network Embeddings
• Compare DNN embeddings and Classic features

Summary:

• The program takes in target filename for T, directory of images as the database B, the feature type, matching method, and the number of images to return.
• The program prints the filename software the top N similar images.