Image Retrieval

Retrieve top k similar images using Haar Cascade, MobileNetV2, KDTree.

Work flow

• Dataset preparation:
  • Using face-detection classifier model Haar Cascade for all images in the dataset.
  • Apply pretrained model MobileNetV2 for extracting all images in the dataset into feature vectors.
  • Save all feature vectors into dictionary and load into .pkl file.
  • Load feature vectors into module.
  • Build KDTree by these feature vectors for optimize indexing step.
• Retrieve image:
  • Upload input image into the module.
  • Apply face-detection classifier model Haar Cascade for input image.
  • Apply pretrained model MobileNetV2 for extracting input image into feature vector.
  • Retrieve top k similar images from KDTree.
  • Display images.

Technology

• Haar Cascade
• MobileNet V2
• KDTree
• Flask

Reference

• Haar Cascade
• MobileNetV2
• KDTree
• Flask