Aggregating binary local descriptors for image retrieval

Abstract

Content-Based Image Retrieval based on local features is computationally expensive because of the complexity of both extraction and matching of local feature. On one hand, the cost for extracting, representing, and comparing local visual descriptors has been dramatically reduced by recently proposed binary local features. On the other hand, aggregation techniques provide a meaningful summarization of all the extracted feature of an image into a single descriptor, allowing us to speed up and scale up the image search. Only a few works have recently mixed together these two research directions, defining aggregation methods for binary local features, in order to leverage on the advantage of both approaches.In this paper, we report an extensive comparison among state-of-the-art aggregation methods applied to binary features. Then, we mathematically formalize the application of Fisher Kernels to Bernoulli Mixture Models. Finally, we investigate the combination of the aggregated binary features with the emerging Convolutional Neural Network (CNN) features. Our results show that aggregation methods on binary features are effective and represent a worthwhile alternative to the direct matching. Moreover, the combination of the CNN with the Fisher Vector (FV) built upon binary features allowed us to obtain a relative improvement over the CNN results that is in line with that recently obtained using the combination of the CNN with the FV built upon SIFTs. The advantage of using the FV built upon binary features is that the extraction process of binary features is about two order of magnitude faster than SIFTs

AI and Custom Marketing for Cultural Initiatives in Regional Projects: Perspectives on the Territorial Integration of Communities with Tourism

“…It is possible to speak of Artificial Intelligence Marketing, thanks to which multiple automated applications and AI-based systems facilitate the traditional process of targeting and customising consumer experiences. More and more companies are moving towards a data-driven approach to make informed decisions to offer customers the most customised experience possible [2].…”

Medoids as a Packing of Orb Image Descriptors

“…To classify images based on descriptors matching, we need to extract features (which can be thousands per image), store them, and perform matching. These stages typically require a lot of computation time and/or memory resources [1]. To overcome these issues, various solutions have been proposed, including the usage of binary descriptors (e.g., BRISK, ORB) instead of float ones (SURF, SIFT) and packing (aggregation) the entire set of image descriptors into smaller quantities.…”

“…To overcome these issues, various solutions have been proposed, including the usage of binary descriptors (e.g., BRISK, ORB) instead of float ones (SURF, SIFT) and packing (aggregation) the entire set of image descriptors into smaller quantities. Different aggregation approaches have been shown to be effective under various conditions [1][2][3][4]. It is worth noting that a simple k-means clustering procedure could be used to quantize descriptors, build a vocabulary of visual words (Bag of Words, BOW), or create some cluster representations of the descriptor sets to compare them more effectively, thus reducing computational efforts.…”

Application of Computer Image Technology in 3D Painting Based on Cloud Computing

“…This paper puts forward that while carrying forward the originality of traditional painting, it also emphasizes the rediscovery and re creation of the borrowing of new media culture [9]. Contemporary Chinese painting creators must go deep into the real national conditions, be close to the joy and joy of the public, correctly grasp the trend of art development, make full use of the advantages of the network age, deeply understand the latest trend of art, and integrate these advanced artistic ideas into painting creation [10].…”