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

Automated analysis of facial images has found eyes still to be a difficult target [90, 96, 97, 125, 215, 221, 229, 230, 248, 360, 509, 692, 709]. The difficulty comes from the diversities in the appearance of eyes due to both structural individuality and motion of eyes, as shown in Fig. 2.1. Past studies have failed to represent these diversities adequately. For example, Tian et al. [616] used a pair of parabolic curves and a circle as a generic eye model, but parabolic curves have too few parameters to represent the complexity of eyelid shape and motion. Statistical models have been deployed to represent such individual differences for the whole eye region [322, 569, 635], but not for subregions, such as the eyelids, due in part to limited variation in training samples.

Meticulously Detailed Eye Region Model

Current automatic facial recognition systems are not robust to changes in illumination, pose, facial expression, and occlusion. In this paper, in order to address the problem of pose change, we propose an algorithm based on a probabilistic approach to face recognition that takes into account the pose difference between probe and gallery images. By using a large facial image database, the CMU PIE database, which contains images of the same set of people taken from many different angles, we have developed a probabilistic model of how facial features change as the pose changes. This model enables us to make our face recognition system more robust to changes of pose in the probe image. The experimental results show that this approach achieves a better recognition rate than conventional face recognition methods over a much larger range of poses. For example, when the gallery contains only images of a frontal face and the probe image varies its pose orientation, the recognition rate shows less than a 10p difference until the probe pose begins to differ by more than 45°, whereas the recognition rate of a PCA-based method begins to drop at differences as small as 10°, and that of a representative commercial system begins to drop at 30°. © 2006 Wiley Periodicals, Inc. Syst Comp Jpn, 37(8): 68–76, 2006; Published online in Wiley InterScience (). DOI 10.1002sscj.20437(This work was done while the first author was visiting the Robotics Institute of Carnegie Mellon)

Multi-subregion-based probabilistic approach to pose-invariant face recognition

: Research supported by this contract includes perception for road following, terrain mapping for off-road navigation, and systems software for building integrated mobile robots. We overview our efforts for the year, and list our publications and personnel, then provide further detail on several of our subprojects. During the past year, this contract has supported research on color vision for road following; 3-D perception for terrain mapping and cross- country mobility; and system building for autonomous navigation. We have demonstrated autonomous navigation on a variety of roads, including single lane dirt, gravel , and paved; and multi-lane roads with and without lane markings. Our perception modules use a variety of techniques for video processing (clustering theory, symbolic feature detection, neural nets), and for range data analysis (landmark navigation, reflectance processing). We have also integrated position-based navigation (INS and GPS), and combinations of all these techniques into mobile robot systems and demonstrations. Our scientific papers this year include a book (Vision and Navigation: the CMU Navlab), three PhD dissertations, and an MS thesis.

/pdf/perception-for-outdoor-navigation-first-year-report-2dymzzb57x.pdf

Perception for Outdoor Navigation. First Year Report

This paper proposes a new image restoration method for phase contrast microscopy as a mean to enhance the quality of images prior to image analysis. Compared to state-of-the-art image restoration algorithms, our method has a more solid theoretical foundation and is orders of magnitude more efficient in computation. We validated the proposed method by applying it to automated muscle myotube detection, a challenging problem that has not been tackled without staining images. Results on 300 phase contrast microscopy images from three different culture conditions demonstrate that the proposed restoration scheme improves myotube detection, and that our approach is far more computationally efficient than previous methods.

/pdf/efficient-phase-contrast-microscopy-restoration-applied-for-4teypd30ve.pdf

Efficient phase contrast microscopy restoration applied for muscle myotube detection.

The findings of a study sponsored by the National Science Foundation and the National Aeronautics and Space Administration's Automation and Robotics Program, prepared for the US government's program evaluating Japanese technology, are summarized. They reveal that the Japanese government, industry, and university leaders have embarked on cooperative projects to develop next-generation robots for space. The goals are to minimize the enormous expense of manned space operations and to spur technology by developing a range of automated machines. Japanese robots are now employed in construction projects on land and underwater, providing experience that may be applicable in space. An overview is given of notable developments to date. >

Takeo Kanade

Papers

Meticulously Detailed Eye Region Model

Multi-subregion-based probabilistic approach to pose-invariant face recognition

Perception for Outdoor Navigation. First Year Report

Efficient phase contrast microscopy restoration applied for muscle myotube detection.

Japan robotics aim for unmanned space exploration