Deep visual words: Improved fisher vector for image classification

Diba, Ali; Pazandeh, Ali Mohammad; Gool, Luc Van

doi:10.23919/mva.2017.7986832

Cited by 9 publications

(12 citation statements)

References 20 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Future works will include the use of the proposed FV coding for covariance matrices descriptors in a hybrid classification architecture which will combine them with convolutional neural networks [17][18][19].…”

Section: Discussionmentioning

confidence: 99%

Fisher Vector Coding for Covariance Matrix Descriptors Based on the Log-Euclidean and Affine Invariant Riemannian Metrics

et al. 2018

View full text Add to dashboard Cite

This paper presents an overview of coding methods used to encode a set of covariance matrices. Starting from a Gaussian mixture model (GMM) adapted to the Log-Euclidean (LE) or affine invariant Riemannian metric, we propose a Fisher Vector (FV) descriptor adapted to each of these metrics: the Log-Euclidean Fisher Vectors (LE FV) and the Riemannian Fisher Vectors (RFV). Some experiments on texture and head pose image classification are conducted to compare these two metrics and to illustrate the potential of these FV-based descriptors compared to state-of-the-art BoW and VLAD-based descriptors. A focus is also applied to illustrate the advantage of using the Fisher information matrix during the derivation of the FV. In addition, finally, some experiments are conducted in order to provide fairer comparison between the different coding strategies. This includes some comparisons between anisotropic and isotropic models, and a estimation performance analysis of the GMM dispersion parameter for covariance matrices of large dimension.

show abstract

Section: Discussionmentioning

confidence: 99%

Fisher Vector Coding for Covariance Matrix Descriptors Based on the Log-Euclidean and Affine Invariant Riemannian Metrics

et al. 2018

View full text Add to dashboard Cite

show abstract

“…Especially, there are many coding methods proposed in the era of BoVW model [5], e.g., hard voting, soft voting, sparse coding, LLC, local coordinate coding, super vector coding, fisher coding, grouping saliency coding, etc. Given the high similarity between BoVW model and BoDVW model in terms of workflow, some of them have been applied in the existing BoDVW methods such as hard voting [12] [13], sparse coding [4], and LLC [10].…”

Section: Related Workmentioning

confidence: 99%

“…There are two methods for extracting deep features in the literature, as shown in Figure 1. The first one (denoted as Ext-by-FC) densely samples multi-scale image patches (e.g., 128 × 128 pixels, 160 × 160 pixels, 192 × 192 pixels) from an image [3] [4] [9] [10]. Each patch is resized to fit the input size (e.g., 224 × 224 pixels for ResNet-50) of an off-the-shelf CNN model, then the output of a deep-level fully-connected layer for each resized patch is regarded as a deep feature.…”

Section: Introductionmentioning

confidence: 99%

Improving Bag-of-Deep-Visual-Words Model via Combining Deep Features With Feature Difference Vectors

Wang

2022

IEEE Access

View full text Add to dashboard Cite

Bag-of-Deep-Visual-Words (BoDVW) model has shown its advantage to Convolutional Neural Network (CNN) model in image classification tasks with a small number of training samples. An essential step in BoDVW model is to extract deep features by using an off-the-shelf CNN model as a feature extractor. Two deep feature extraction methods have been raised in recent years. The first method densely samples multi-scale image patches and then converts them into deep features via a deep-level fullyconnected layer.The second method uses the output of a deep-level convolutional layer or pooling layer as the source of deep features. By contrast, the second method is much more efficient. However, it performs worse than the first method in classification accuracy. The reason is that deep features extracted by the second method are yielded in receptive fields of a single size. To make BoDVW model have high feature extraction efficiency and high classification accuracy, we propose enhancing deep features extracted by the second method at low added computation costs by supplementing the information obtained from receptive fields of different sizes. Concretely, we raise a novel feature named "feature difference (FD) vector" in this article. It can roughly preserve the information of multiple deep features extracted by the convolutional layers of different receptive field sizes. Each deep feature is enhanced by combining an FD vector to form a combined feature. The image representation vector of an image is generated using the combined features extracted from it. Our experimental results on three public datasets (15-Scenes, TF-Flowers, and NWPU-RESISC45) show that our method can avoid the high computation costs of the first method and achieve comparable results to the first method, which exhibits the effectiveness of our method.

show abstract

“…Diba et al. [41] proposed an iterative framework to obtained improved FV features for the image classification task. Yongsheng Pan et al.…”

Section: Related Workmentioning

confidence: 99%

A graph convolutional neural network model with Fisher vector encoding and channel‐wise spatial‐temporal aggregation for skeleton‐based action recognition

Tang

Wang

et al. 2022

IET Image Processing

View full text Add to dashboard Cite

Skeleton‐based action recognition is an inspired yet challenging task in computer vision. Recently, the latest graph convolutional network (GCN), which generalises well‐established convolutional neural networks to non‐Euclidean structures, is proven to be highly successful for action recognition from body skeleton data. However, the GCN architecture has not been fully studied. In this work, a Fisher vector (FV) encoding based GCN architecture (FV‐GCN) is proposed, which exceeds the limitations of existing GCN‐based methods by combining the GCN model with FV encoding. A channel‐wise spatial–temporal aggregation function to preserve spatial–temporal information in the whole action clip and integrate it into the FV‐GCN architecture is also presented. Since FV is different from the GCN structure, this hybrid architecture that incorporates the advantages of both algorithms can discover complementary information of feature representation effectively. On two challenging human action datasets, kinetics, and NTU‐RGBD, improved performance is demonstrated over the baseline method, and the FV‐GCN is better or comparable to some state‐of‐the‐art methods.

show abstract

Deep visual words: Improved fisher vector for image classification

Cited by 9 publications

References 20 publications

Fisher Vector Coding for Covariance Matrix Descriptors Based on the Log-Euclidean and Affine Invariant Riemannian Metrics

Fisher Vector Coding for Covariance Matrix Descriptors Based on the Log-Euclidean and Affine Invariant Riemannian Metrics

Improving Bag-of-Deep-Visual-Words Model via Combining Deep Features With Feature Difference Vectors

A graph convolutional neural network model with Fisher vector encoding and channel‐wise spatial‐temporal aggregation for skeleton‐based action recognition

Contact Info

Product

Resources

About