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

Phase contrast and differential interference contrast, which are used to capture microscopy images of living cells, contain a few artifacts such as halo and shadow-cast effect. Removing these artifacts from microscopy images facilitates automated microscopy image analysis, such as the cell segmentation that is a critical step in cell tracking systems. We propose a restoration algorithm based on the microscopy imaging model and consider temporal consistency when restoring time-lapse microscopy image sequences. The artifact-free microscopy images are restored by minimizing a regularized quadratic cost function that is adaptable to input image properties. Our method achieves high segmentation accuracy and low computational cost compared to the previous methods.

/pdf/restoring-artifact-free-microscopy-image-sequences-3nhao1c57j.pdf

Restoring artifact-free microscopy image sequences

Coordination, Organizations, Institutions and Norms in Agent Systems IV

The authors present a model-based object recognition system for specular objects. Objects with specular surfaces present a problem for computer vision. Simulating object appearances by using the sensor model, and the object model allows us to predict specular features, and to analyze the detectability and reliability of each feature. The system generates a set of aspects of the object. By precompiling the aspects with the feature detectability and the feature reliability, the system prepares adaptable matching templates. At the runtime, an input image is first classified into a few candidate aspects. A deformable template matching finds the best match among them. This method is applicable to multiple objects simply by changing object and sensor models. Experimental results using two kinds of objects and sensors are presented: a TV image of a shiny object and a synthetic aperture radar (SAR) image of an airplane. The results show the flexibility of the proposed model based approach. >

Model based recognition of specular objects using sensor models

The visual medium evolved from early paintings to the realistic paintings of the classical era to photographs. The medium of moving imagery started with motion pictures. Television and video recording advanced it to show action “live” or capture and playback later. In all of the above media, the view of the scene is determined at the transcription time, independent of the viewer.

Virtual-ized Reality: Being Mobile in a Visual Scene

: Video-rate z keying is a new image keying method for merging real and synthetic images in real time. In visual media communication and display, it is often necessary to merge video signals from a real camera and a synthetic video produced by computer graphics. A standard technique for such a purpose is chroma keying which is used, for example, in TV weather reports. Chroma keying, however, simply puts real world objects in the foreground of the synthetic image, and cannot deal with situation where real and synthetic objects occlude each other. The z key method we present merges real and virtual world images in a more flexible way. The z key uses pixel-by-pixel depth information in the form of a depth map as a switch. For each pixel, the z key switch compares the pixel depth values of two images, and routes the color value of the foreground image that is nearer to the camera for the merged output image. The result of this pixel-by-pixel switching is that real and virtual objects can occlude each other correctly depending on their geometrical relationships. The critical capability for realizing such video-rate z keying is video-rate pixel-by-pixel depth mapping of a real scene. We have developed a video-rate stereo machine which can produce 200 x 200 depth images at video rate. With this machine, merging a real scene with a synthetic scene by means of z keying in real-time has been demonstrated; a real person walks around in a synthetic room with correct relationships with virtual objects in the room at the rate of 15 frames/sec.

/pdf/video-rate-z-keying-a-new-method-for-merging-images-481jl4rccw.pdf

Takeo Kanade

Papers

Restoring artifact-free microscopy image sequences

Coordination, Organizations, Institutions and Norms in Agent Systems IV

Model based recognition of specular objects using sensor models

Virtual-ized Reality: Being Mobile in a Visual Scene

Video-Rate Z Keying: A New Method for Merging Images