Learning features for offline handwritten signature verification using deep convolutional neural networks
Citations
184 citations
Cites background from "Learning features for offline handw..."
...However, deep learning [4, 11, 46, 109, 110, 139, 145, 220, 247, 275] seems to be one of the hot topics in ASV....
[...]
...Such work was extended in [110] by a novel formulation of the problem that includes applying a knowledge of skilled forgeries during the feature learning process....
[...]
...- Statistical models Neural Networks (NNs) [144] and Deep Learning (Recurrent Neural Networks (RNNs) [4, 146, 247],*, Convolutional Neural Networks (CNNs) [268],* [11, 46, 109, 110, 139],** Deep Neural Networks (DNN) [220],** Deep Multitask Metric Learning (DMML) [240],** DCGANs [275]**), Hidden Markov Models (HMMs) [15, 72, 157, 251],* [48, 74],** Support Vector Machine (SVM) [103],* [54, 56, 76, 104, 198, 274],** Random Forest [203],* ....
[...]
135 citations
Cites background or methods from "Learning features for offline handw..."
...6 WD [28] Feature learning (SVM) 12 - 4....
[...]
...94 WD [28] Feature learning (SVM) 12 3....
[...]
...Recent studies approach the problem from a representation learning perspective [27], [28], [50], [68]: instead of designing feature extractors for the task, these methods rely on learning feature representations directly from signature images....
[...]
...work [28], the authors also proposed a multi-task framework, where the CNN is trained with both genuine signatures and skilled forgeries, optimizing to jointly discriminate between users, and discriminate between genuine signatures and forgeries....
[...]
...Other authors use a fixed frame size (width and height), and center the signature in this frame [49], [28]....
[...]
132 citations
88 citations
62 citations
References
73,978 citations
55,235 citations
"Learning features for offline handw..." refers methods in this paper
...gued in [37], these pre-trained models offer a strong baseline for Computer Vision tasks. We used two pre-trained models3, namely Caffenet (Caffe reference network, based on AlexNet [25]), and VGG-19 [38]. 3https://github.com/BVLC/caffe/wiki/Model-Zoo 18 We used these networks to extract the feature representations ˚(X) for signatures, and followed the same protocol for training Writing-Dependent clas...
[...]
49,914 citations
[...]
46,982 citations
"Learning features for offline handw..." refers methods in this paper
... images. Methods based on learning multiple levels of representation have shown to be very effective to process natural data, especially in computer vision and natural language processing [21], [22], [23]. The intuition is to use such methods to learn multiple intermediate representations of the input, in layers, in order to better represent a given problem. In a classification task, the higher layers ...
[...]
...ut that are important for classification, while disregarding irrelevant variations [23]. In particular, Convolutional Neural Networks (CNNs) [24] have been used to achieve state-of-the-art performance [23] in many computer vision tasks [25], [26]. These models use local connections and shared weights, taking advantage of the spatial correlations of pixels in images by learning and using the same filters...
[...]
40,257 citations