Document image defect models

By Henry S. Baird. 1995

Subfields: Computer Vision and Pattern Recognition

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

Citations in the corpus

Toward Causal Representation Learning (2021, background): “Distribution shifts may also arise from simple corruptions that are common in real-world data collection pipelines [10], [107], [129], [170], [206].”
Deep learning in neural networks: An overview (2014, methodology): “A BP-trained (LeCun et al., 1989) CNN (Sections 5.4, 5.8) set a newMNIST record of 0.4% (Simard, Steinkraus, & Platt, 2003), using training pattern deformations (Baird, 1990) but no unsupervised pre-training (Sections 5.7, 5.10, 5.15).”
Understanding Data Augmentation for Classification: When to Warp? (2016, background): “The concept of data warping (for neural networks) can possibly be attributed to [11] in creating a model of character distortions that could occur in the printing (or handwriting) process, the model described deformations and defects needed to generate synthetic character examples.”
Towards Causal Representation Learning (2021, methodology): “is to rely on data augmentation to increase the diversity of the data by “augmenting” it through a certain type of artificially generated interventions [9, 234, 140].”
Learning Data Augmentation Strategies for Object Detection (2019, neutral): “Right: Mean average precision for RetinaNet [24] with a ResNet-50 backbone on COCO [25] with and without learned augmentation policy (red and black, respectively). augmentation strategies [3, 42, 21].”
AutoAugment: Learning Augmentation Strategies From Data (2019, methodology): “Data augmentation is an effective technique to increase both the amount and diversity of data by randomly “augmenting” it [3, 54, 29]; in the image domain, common augmentations include translating the image by a few pixels, or flipping the image horizontally.”
Dreaming More Data: Class-dependent Distributions over Diffeomorphisms for Learned Data Augmentation (2015, methodology): “This form of data augmentation is a rather old idea [6, 42], which is still considered part of “best practices” [43] for many tasks.”
Unsupervised Feature Learning and Deep Learning: A Review and New Perspectives (2012, methodology): “While being an old approach (Baird, 1990; Poggio and Vetter, 1992), it has been recently applied with great success in the work of Ciresan et al. (2010) who used an efﬁcient GPU implementation ( 40 (cid:2) speedup) to train a standard but large deep multilayer Perceptron on deformed MNIST digits.”
Adaptive document image binarization (2000, background): “For example, a letter connectivity must be maintained for optical character recognition and textual compression [2].”
Learning over Sets using Kernel Principal Angles (2003, methodology): “Keywords: Kernel Machines, Large margin classifiers, Canonical Correlation Analysis.”
An overview of character recognition focused on off-line handwriting (2001, background): “There is very little work on modeling the noise introduced by optical distortion, like speckle, skew, and blur [11], [99], [151].”
My First Deep Learning System of 1991 + Deep Learning Timeline 1962-2013 (2013, methodology): “Our team ﬁrst showed that good old backpropagation (TL 1.2) on GPUs (with training pattern distortions [6, 86] but without any unsupervised pre-training) can actually achieve a three times better result of 0.35% [17] - back then, a world record (a previous standard net achieved 0.7% [86]; a…”
Deep learning for time series classification: a review (2018, methodology): “Another popular technique is data augmentation, which tackles the problem of overfitting a small dataset by increasing the number of training instances (Baird 1992).”
Stacked Denoising Autoencoders: Learning Useful Representations in a Deep Network with a Local Denoising Criterion (2010, background): “…performance of a classiﬁer (neural network or other), which consists in augmenting the original training set with additional distorted inputs, either explicitly (Baird, 1990; Poggio and Vetter, 1992) or virtually through a modiﬁed algorithm (Simard et al., 1992; Sch ¨ olkopf et al., 1996).”
Dash: Semi-Supervised Learning with Dynamic Thresholding (2021, methodology): “With the image data augmentation technique, consistency regularization uses unlabeled data [Baird, 1992, Schmidhuber, 2015] based on the condition that the model predictions between diﬀerent perturbed versions of the same image are similar.”
Survey of the State of the Art in Human Language Technology (2010, background): “, and Baird, H. S. (1987). On the Recognition of Printed Characters of Any Font and Size.”