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

Registration of medical images enables quantitative study of anatomical differences between populations, as well as detection of abnormal variations indicative of pathologies. However inherent anatomical variabilities between individuals and possible pathologies make registration difficult. This paper presents a bootstrap strategy for characterizing non-pathological variations in human brain anatomy, as well its application to achieve accurate 3-D deformable registration. Inherent anatomical variations are initially extracted by deformably registering training data with an expert-segmented 3-D image, a digital brain atlas. Statistical properties of the density and geometric variations in brain anatomy are extracted and encoded into the atlas to build a statistical atlas. These statistics are then used as prior knowledge to guide the deformation process. A bootstrap loop is formed by registering the statistical atlas to larger training sets as more data becomes available, so as to ensure more robust knowledge extraction, and to achieve more precise registration. Compared to an algorithm with no knowledge guidance, registration using the statistical atlas reduces the overall error by 34%.

Bootstrap a statistical brain atlas

This paper introduces the Spotting by Association method for video analysis, which is a novel method to detect video segments with typical semantics. Video data contains various kinds of information by means of continuous images, natural language, and sound. For use in a Digital Library, it is essential to segment the video data into meaningful pieces. To detect meaningful segments, we should associate data from each modality, including video, language, and sound. For this purpose, we propose a new method for segment spotting by making correspondences between image clues detected by image analysis and language clues created by natural language analysis. As a result, relevant video segments with sufficient information in every modMity are obtained. We applied our method to closed-captioned CNN Headline News. Video segments with important situations, that is a speech, meeting, or visit, are detected fairly well.

https://www.aaai.org/Papers/Symposia/Spring/1997/SS-97-03/SS97-03-019.pdf

Spotting by Association in News Video

Time-lapse live cell imaging has been increasingly employed by biological and biomedical researchers to understand the underlying mechanisms in cell physiology and development by investigating behavior of cells. This trend has led to a huge amount of image data, the analysis of which becomes a bottleneck in related research. Consequently, how to efficiently analyze the data is emerging as one of the major challenges in the fields. 
Computer vision analysis of non-fluorescent microscopy images, representatively phase-contrast microscopy images, promises to realize a long-term monitoring of live cell behavior with minimal perturbation and human intervention. To take a step forward to such a system, this thesis proposes computer vision algorithms that monitor cell growth, migration, and differentiation by detecting three cellular events—mitosis (cell division), apoptosis (programmed cell death), and differentiation—and tracking individual cells. Among the cellular events, to the best our knowledge, apoptosis and a certain type of differentiation, namely muscle myotubes, have never been detected without fluorescent labeling. We address these challenging problems by developing computer vision algorithms adopting phase contrast microscopy. We also significantly improve the accuracy of mitosis detection and cell tracking in phase contrast microscopy over previous methods, particularly under non-trivial conditions, such as high cell density or confluence. We demonstrate the usefulness of our methods in biological research by analyzing cell behavior in scratch wound healing assays. The automated system that we are pursuing would lead to a new paradigm of biological research by enabling quantitative and individualized assessment in behavior of a large population of intact cells.

Takeo Kanade

Papers

Bootstrap a statistical brain atlas

Spotting by Association in News Video

Toward an automated system for the analysis of cell behavior: cellular event detection and cell tracking in time-lapse live cell microscopy

Stereo vision with arbitrary camera arrangement and with camera calibration

Reference consistent reconstruction of 3D cloth surface