Aggregation of Deep Local Features using VLAD and Classification using R Forest
| Autor: | Ankit Anand, Rajat Nigam, S Natarajan, A. Vinay, K. N. Balasubramanya Murthy, Abhijay Gupta, Harsh Garg |
|---|---|
| Rok vydání: | 2018 |
| Předmět: |
0301 basic medicine
Contextual image classification Computer science business.industry ComputingMethodologies_IMAGEPROCESSINGANDCOMPUTERVISION Pattern recognition 02 engineering and technology Facial recognition system Ensemble learning Convolutional neural network Image (mathematics) Random forest 03 medical and health sciences ComputingMethodologies_PATTERNRECOGNITION 030104 developmental biology Computer Science::Computer Vision and Pattern Recognition 0202 electrical engineering electronic engineering information engineering Feature (machine learning) General Earth and Planetary Sciences 020201 artificial intelligence & image processing Artificial intelligence business General Environmental Science |
| Zdroj: | Procedia Computer Science. 143:998-1006 |
| ISSN: | 1877-0509 |
| DOI: | 10.1016/j.procs.2018.10.334 |
| Popis: | The paper proposes an efficient and accurate model for face recognition using an attentive local feature descriptor extracted from Convolutional Neural Network referred to as DEep Local Feature (DELF). The algorithm mentioned formerly is used for extracting descriptors from the images using a fully convolutional network which are trained with weak supervision and using image level classes, neglecting the usage of patch and object level annotations. The physical characteristics such as colour, texture, etc are represented in the form of 40 dimensional vectors using DELF. Further, such descriptors are quantized to represent them into the compact form using Vector of Locally Aggregated Descriptors and Fisher kernels. Subsequently, such vectors are used for multi-class image classification using ensemble learning methods including Rotation Forest and Random Forests. Comparative study between both the classifiers and feature aggregation methods are performed and tabulated in the paper. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|