Multilayer neural networks trained with the back-propagation algorithm constitute the best example of a successful gradient based learning technique. Given an appropriate network architecture, gradient-based learning algorithms can be used to synthesize a complex decision surface that can classify high-dimensional patterns, such as handwritten characters, with minimal preprocessing. This paper reviews various methods applied to handwritten character recognition and compares them on a standard handwritten digit recognition task. Convolutional neural networks, which are specifically designed to deal with the variability of 2D shapes, are shown to outperform all other techniques. Real-life document recognition systems are composed of multiple modules including field extraction, segmentation recognition, and language modeling. A new learning paradigm, called graph transformer networks (GTN), allows such multimodule systems to be trained globally using gradient-based methods so as to minimize an overall performance measure. Two systems for online handwriting recognition are described. Experiments demonstrate the advantage of global training, and the flexibility of graph transformer networks. A graph transformer network for reading a bank cheque is also described. It uses convolutional neural network character recognizers combined with global training techniques to provide record accuracy on business and personal cheques. It is deployed commercially and reads several million cheques per day.

Gradient-based learning applied to document recognition

We propose a deep convolutional neural network architecture codenamed Inception that achieves the new state of the art for classification and detection in the ImageNet Large-Scale Visual Recognition Challenge 2014 (ILSVRC14). The main hallmark of this architecture is the improved utilization of the computing resources inside the network. By a carefully crafted design, we increased the depth and width of the network while keeping the computational budget constant. To optimize quality, the architectural decisions were based on the Hebbian principle and the intuition of multi-scale processing. One particular incarnation used in our submission for ILSVRC14 is called GoogLeNet, a 22 layers deep network, the quality of which is assessed in the context of classification and detection.

/pdf/going-deeper-with-convolutions-1yobw2o2ds.pdf

Going deeper with convolutions

There is, I think, something ethereal about i —the square root of minus one. I remember first hearing about it at school. It seemed an odd beast at that time—an intruder hovering on the edge of reality.

Usually familiarity dulls this sense of the bizarre, but in the case of i it was the reverse: over the years the sense of its surreal nature intensified. It seemed that it was impossible to write mathematics that described the real world in …

I and i

We study the question of feature sets for robust visual object recognition; adopting linear SVM based human detection as a test case. After reviewing existing edge and gradient based descriptors, we show experimentally that grids of histograms of oriented gradient (HOG) descriptors significantly outperform existing feature sets for human detection. We study the influence of each stage of the computation on performance, concluding that fine-scale gradients, fine orientation binning, relatively coarse spatial binning, and high-quality local contrast normalization in overlapping descriptor blocks are all important for good results. The new approach gives near-perfect separation on the original MIT pedestrian database, so we introduce a more challenging dataset containing over 1800 annotated human images with a large range of pose variations and backgrounds.

/pdf/histograms-of-oriented-gradients-for-human-detection-3nteinziuq.pdf

Histograms of oriented gradients for human detection

Object detection performance, as measured on the canonical PASCAL VOC dataset, has plateaued in the last few years. The best-performing methods are complex ensemble systems that typically combine multiple low-level image features with high-level context. In this paper, we propose a simple and scalable detection algorithm that improves mean average precision (mAP) by more than 30% relative to the previous best result on VOC 2012 -- achieving a mAP of 53.3%. Our approach combines two key insights: (1) one can apply high-capacity convolutional neural networks (CNNs) to bottom-up region proposals in order to localize and segment objects and (2) when labeled training data is scarce, supervised pre-training for an auxiliary task, followed by domain-specific fine-tuning, yields a significant performance boost. Since we combine region proposals with CNNs, we call our method R-CNN: Regions with CNN features. We also present experiments that provide insight into what the network learns, revealing a rich hierarchy of image features. Source code for the complete system is available at http://www.cs.berkeley.edu/~rbg/rcnn.

/pdf/rich-feature-hierarchies-for-accurate-object-detection-and-22fvjlggyg.pdf

Rich Feature Hierarchies for Accurate Object Detection and Semantic Segmentation

Mackworth's gradient space has proved to be a useful tool for image understanding. However, descriptions of its important properties have been somewhat scattered in the literature. The fundamental properties of the gradient space under orthography and perspective, and for curved surfaces, are developed and summarized. While largely a recounting of previously published results, there are a number of new observations, particularly concerning the gradient space and perspective projection. In addition, the definition and use of vector gradients as well as surface gradients provides concise notation for several results. The properties explored include the orthographic and perspective projections themselves; the definition of gradients; the gradient space consequences of vectors (edges) belonging to one or more surfaces, and of several vectors being contained on a single surface; and the relationships between vanishing points, vanishing lines, and the gradient space. The paper is intended as a study guide for learning about the gradient space, as well as a reference for researchers working with gradient space.

Gradient space under orthography and perspective

Three-dimensional (3D) kinematic models are widely-used in video-based figure tracking. We show that these models can suffer from singularities when motion is directed along the viewing axis of a single camera. The single camera case is important because it arises in many interesting applications, such as motion capture from movie footage, video surveillance, and vision-based user-interfaces.We describe a novel two-dimensional scaled prismatic model (SPM) for figure registration. In contrast to 3D kinematic models, the SPM has fewer singularity problems and does not require detailed knowledge of the 3D kinematics. We fully characterize the singularities in the SPM and demonstrate tracking through singularities using synthetic and real examples.We demonstrate the application of our model to motion capture from movies. Fred Astaire is tracked in a clip from the film “Shall We Dance”. We also present the use of monocular hand tracking in a 3D user-interface. These results demonstrate the benefits of the SPM ...

/pdf/ambiguities-in-visual-tracking-of-articulated-objects-using-fraqmitw04.pdf

Ambiguities in Visual Tracking of Articulated Objects Using Two- and Three-Dimensional Models:

We address the problem of 6D pose estimation of a textureless and shiny object from single-view 2D images, for a bin-picking task. For a textureless object like a mechanical part, conventional visual feature matching usually fails due to the absence of rich texture features. Hierarchical template matching assumes that few templates can cover all object appearances. However, the appearance of a shiny object largely depends on its pose and illumination. Furthermore, in a bin-picking task, we must cope with partial occlusions, shadows, and inter-reflections.

/pdf/6d-pose-estimation-of-textureless-shiny-objects-using-random-25q0uyeqd1.pdf

6D pose estimation of textureless shiny objects using random ferns for bin-picking

We describe the current state of the 3-D Mosaic project, whose goal is to incrementally acquire a 3-D model of a complex urban scene from images. The notion of incremental acquisition arises from the observations that 1) single images contain only parfial information about a scene, 2) complex images are difficult to fully interpret, and 3) different features of a given scene tend to be easier to extract in different images because of differences in viewpoint and lighting conditions. In our approach, multiple images of the scene are sequentially analyzed so as to incrementaly construct the model. Each new image provides information which refines the model. We describe some experiments toward this end. Our method of extracting 3-D shape information from the images is stereo analysis. Because we are dealing with urban scenes, a junction-based matching technique proves very useful. This technique produces rather sparse wire-frame descriptions of the scene. A reasoning system that relies on task-specific knowledge generates an approximate model of the scene from the stereo output. Gray scale information is also acquired for the faces in the model. Finally, we describe an experiment in combining two views of the scene to obtain a rermed model.

Incremental acquisition of a three-dimensional scene model from images

Recursive segmentation of an image into regions using histograms is one of the most widely used techniques for image segmentation. At CMU, several versions of a region segmentation program have been developed based on this technique (Ohlander, Price, Shafer and Kanade). Based on these experiences, this paper discusses issues of recursive region segmentation in the context of PHOENIX, the newest version of region segmentation program, running on a VAX 11/780 under UNIX. The issues discussed in this paper include: Image features to be used in histogramming; comparison of the algorithm with other techniques; important improvements made in PHOENIX over its predecessor (Ohlander and Price); and some inherent problems in histogram-based segmentation together with suggestions for minimizing them. PHOENIX is being incorporated into the ARPA Image Understanding Testbed, under construction at SRI International.

Takeo Kanade

Papers

Gradient space under orthography and perspective

Ambiguities in Visual Tracking of Articulated Objects Using Two- and Three-Dimensional Models:

6D pose estimation of textureless shiny objects using random ferns for bin-picking

Incremental acquisition of a three-dimensional scene model from images

Recursive region segmentation by analysis of histograms