Feature Extraction, Dimensionality Reduction and Query Processing on an Image Database.

The project uses dataset associated with the following publication:
Mahmoud Afifi. “11K Hands: Gender recognition and biometric identification using a large dataset of hand images.” M. Multimed Tools Appl (2019) 78: 20835

• Feature-Extraction:
  • Extracting features using from the 11k hand images using the following method:
    • Color Moments
    • Local Binary Pattern
    • Histograms of oriented gradients
    • Scale-invariant feature transform
  • Implemented a program to extract k similar images, provided a image-id using the extracted feature descriptors and using Cosine Similarity as a distance measure.

• Impletented a program to identify latent semantics using the following 4 methods:
  • Pricipal Component Analysis (PCA)
  • Singular Value Decomposition (SVD)
  • Non-negative Matrix Factorization (NMF)
  • Latent Dirichlet Analysis (LDA)
• Implemented a program to list most related m images in the k dimanesional latent space.
• Implemented a program which (a) lets the user to chose among one of the four feature models and (b) given one of the labels
  • left-hand
  • right-hand
  • dorsal
  • palmar
  • with accessories
  • without accessories
  • male
  • female
    identifies (and lists) k latent semantics for images with the corresponding metadata using (c) one of the following techniques chosen by the user
• Implement a program which given ( a subject ID, identifies and visualizes the most related 3 subjects).

• Implemented a program which, given a folder with dorsal/palmar labeled images and for a user supplied c,
  • computes c clusters associated with dorsal-hand images (visualize the resulting image clusters),
  • computes c clusters associated with palmar-hand images (visualize the resulting image clusters),
    and, given a folder with unlabeled images, the system labels them as dorsal-hand vs palmar-hand using descriptors of these clusters.
• Implemented Personalised Page Rank (PPR) Algorithm to detect and visualise K most dominant images for a given image-id.
• Implemented SVM, Decision-tree and PPR based Algorithm to classify a set of images as dorsal-hand vs palmar-hand .
• Implemented Locality Sensitive Hashing based search algorithm to visualise most similar images for a given image-id.
• Impleted a Relevant Feedback System.