Efficient large-scale image search with a vocabulary tree
The task of searching and recognizing objects in images has become an important research topic in the area of image processing and computer vision. Looking for similar images in large datasets given an input query and responding as fast as possible is a very challenging task. In this work the Bag of...
Guardado en:
| Autores principales: | , , |
|---|---|
| Formato: | JOUR |
| Materias: | |
| Acceso en línea: | http://hdl.handle.net/20.500.12110/paper_21051232_v8_n_p71_Uriza |
| Aporte de: |
| Sumario: | The task of searching and recognizing objects in images has become an important research topic in the area of image processing and computer vision. Looking for similar images in large datasets given an input query and responding as fast as possible is a very challenging task. In this work the Bag of Features approach is studied, and an implementation of the visual vocabulary tree method from Nistér and Stewénius is presented. Images are described using local invariant descriptor techniques and then indexed in a database using an inverted index for further queries. The descriptors are quantized according to a visual vocabulary, creating sparse vectors, which allows to compute very efficiently, for each query, a ranking of similarity for indexed images. The performance of the method is analyzed varying different factors, such as the parameters for the vocabulary tree construction, different techniques of local descriptors extraction and dimensionality reduction with PCA. It can be observed that the retrieval performance increases with a richer vocabulary and decays very slowly as the size of the dataset grows. © 2018 IPOL and the authors CC-BY-NC-SA. |
|---|