Deep Feature Pyramid Hashing for Efficient Image Retrieval

Thanks to the success of deep learning, hashing has recently evolved as a leading method for large-scale image retrieval. Most existing methods use last layer extract semantic information from input image. However, these have deficiencies because features extracted lack local information, which might impact global system’s performance. To this end, Deep Feature Pyramid Hashing DFPH is proposed in study, can fully utilize images’ multi-level visual and information. Our architecture applies new feature pyramid network designed VGG-19 model, so model becomes able learn hash codes various scales then fuse them create final binary codes. The experimental results performed on two widely used retrieval datasets demonstrate superiority our method.