Abstract

Robust segmentation of ballistic strokes fromonline handwritten traces is critical in parameter estimationof stroke based models for applications such as recognition,synthesis, and writer identification. In this paper we propose anew method for segmenting ballistic strokes from online hand-writing. Traditional methods of ballistic stroke segmentationrely on detection of local minima of pen speed. Unfortunately,this approach is highly sensitive to noise, in sensing and inboth spatial and temporal dimensions. We decompose theproblem into two steps, where the spatial noise is filteredout in the first step. The ballistic stroke boundaries are thendetected at the local curvature maxima, which we show to beinvariant to temporal sampling noise. We also propose a bag-of-strokes representation based on ballistic stroke segmentationfor online character recognition that improves the state-of-the-art recognition accuracies on multiple datasets

Abstract

We present a novel method for reducing the effectsof ink-bleed in handwritten documents. We go beyond the existing works on ink bleed detection and removal. We consider each pixel in a document as a result of combination of foreground,ink-bleed and background. We carry of a decomposition of the document image into separate foreground ink, ink-bleed, and background layers. We propose an efficient MRF formulation to achieve this separation. Degradation model for the ink and paper is proposed. The ability to extract the contributions of the three components to each pixel allows us to recover finer details of the writing. Quantitative and qualitative results on a set of historic manuscripts as well as synthetically generated documents demonstrate the effectiveness of our approach

Abstract

We aim to learn local orientation field patterns in fingerprints and correct distorted field patterns in noisy fingerprint images. This is formulated as a learning problem and achieved using two continuous restricted Boltzmann machines. The learnt orientation fields are then used in conjunction with traditional Gabor based algorithms for fingerprint enhancement. Orientation fields extracted by gradient-based methods are local, and do not consider neighboring orientations. If some amount of noise is present in a fingerprint, then these methods perform poorly when enhancing the image, affecting fingerprint matching. This paper presents a method to correct the resulting noisy regions over patches of the fingerprint by training two continuous restricted Boltzmann machines. The continuous RBMs are trained with clean fingerprint images and applied to overlapping patches of the input fingerprint. Experimental results show that one can successfully restore patches of noisy fingerprint images.

Abstract

Binarization of text in natural scene images is a challenging task due to the variations in color, size, and font of the text and the results are often affected by complex back-grounds, different lighting conditions, shadows and reflections.A robust solution to this problem can significantly enhance the accuracy of scene text recognition algorithms leading to a variety of applications such as scene understanding, automatic localization and navigation, and image retrieval. In this paper,we propose a method to extract and binarize text from images that contains complex background. We use an Independent Component Analysis (ICA) based technique to map out the text region, which is inherently uniform in nature, while removing shadows, specularity and reflections, which are included in the background. The technique identifies the text regions from the components extracted by ICA using a global thresholding method to isolate the foreground text. We show the results of our algorithm on some of the most complex word images from the ICDAR 2003 Robust Word Recognition Dataset and compare with previously reported methods

Abstract

Sparse representations have emerged as a powerful approach for encoding images in a large class of machine recognition problems including face recognition. These methods rely on the use of an over-complete basis set for representing an image. This often assumes the availability of a large number of labeled training images, especially for high dimensional data. In many practical problems, the number of labeled training samples are very limited leading to significant degradations in classification performance. To address the problem of lack of training samples, we propose a semi-supervised algorithm that labels the unlabeled samples through a multi-stage label propagation combined with sparse representation. In this representation, each image is decomposed as a linear combination of its nearest basis images, which has the advantage of both locality and sparsity.Extensive experiments on publicly available face databases show that the results are significantly better compared to state-of-the-art face recognition methods in semi-supervised setting and are on par with fully supervised techniques

Abstract

Reconstructing geometric models of relief carv-ings are of great importance in preserving cultural heritages digitally. In case of reliefs, using laser scanners and structured lighting techniques is not always feasible or are very expensive given the uncontrolled environment. Single image shape from shading is an under-constrained problem that tries to solve for the surface normals given the intensity image. Various constraints are used to make the problem tractable. To avoid the uncontrolled lighting, we use a pair of images with and without the flash and compute an image under a known illumination.This image is used as an input to the shape reconstruction algorithms. We present techniques that try to reconstruct theshape from relief images using the prior information learned from examples. We learn the variations in geometric shape corresponding to image appearances under different lighting conditions using sparse representations. Given a new image,we estimate the most appropriate shape that will result in the given appearance under the specified lighting conditions. We integrate the prior with the normals computed from reflectance equation in a MAP framework. We test our approach on relief images and compare them with the state-of-the-art shape from shading algorithms.

Abstract

n this paper we present a max-flow min-cut based salient object detection in 3D point cloud that results from Struc-ture from Motion (SfM) pipeline. The SfM pipeline generates noisy point cloud due to the unwanted scenes captured along with the object in the image dataset of SfM. The background points being sparse and not meaningful, it becomes necessary to remove them. Hence, any further processes (like surface reconstruction)utilizing the cleaned up model will have no hinderance from the noise removed. We present a novel approach where the camera centers are used to segment out the salient object. The algorithm is completely autonomous and does not need any user input. We test our proposed method on Indian historical models reconstructed through SfM. We evaluate the results in terms of selectivity and specificity.

Abstract

his paper addresses the problem of reconstruction of specular surfaces using a combination of Dynamic Program­ming and Markov Random Fields formulation. Unlike traditional methods that require the exact position of environment points to be known, our method requires only the relative position of the environment points to be known for computing approximate normals and infer shape from them. We present an approach which estimates the depth from dynamic programming routine and MRF stereo matching and use MRF optimization to fuse the results to get the robust estimate of shape. We used smooth color gradient image as our environment texture so that shape can be recovered using just a single shot. We evaluate our method using synthetic experiments on 3D models like Stanford bunny and show the real experiment results on golden statue and silver coated statue.

Abstract

3D textures are often described by parametric functions for each pixel, that models the variation in its appearance with respect to varying lighting direction. However, parametric models such as Polynomial Texture Maps (PTMs) tend to smoothen the changes in appearance. We propose a technique to effectively model natural material surfaces and their interactions with changing light conditions. We show that the direct and global components of the image have different nature, and when modeled separately, leads to a more accurate and compact model of the 3D surface texture. For a given lighting position, both components are computed separately and combined to render a new image. This method models sharp shadows and specularities, while preserving the structural relief and surface color. Thus rendered image have enhanced photorealism as compared to images rendered by existing single pixel models such as PTMs.

Abstract

Biometric identification often involves explicit comparison of a probe against each template stored in a database.This process becomes extremely time-consuming as the size of the database increases. Filtering approaches use a lightweight comparison to select a smaller set of candidate templates from the database for explicit comparison. However,most existing filtering schemes use specific features that are hand-crafted for the biometric trait at each stage of the filtering. In this work, we explore the effectiveness of weak features in a cascade for filtering fingerprint databases. We start with a set of potential indexing features computed from minutiae triplets and minutiae quadruplets. Each stage of filtering consists of projecting the probe onto a specific line and the removal of database samples outside a window around the probe. The critical problem in this process is the selection of lines for projection at each stage of the filtering. We show that by using a set of random lines and the proposed fitness function, one can achieve better results that optimization methods such as PCA or LDA. Experimental results show that using an ensemble of projections we can reduce the penetration to26%at a hit rate of99%. As each stage of the cascade is extremely fast, and filtering is progressive along the cascade, one can terminate the cascade at any point to achieve the desired performance. One can also combine this method with other indexing methods to improve the overall accuracy and speed. We present de-tailed experimental results on various aspects of the process on the FVC 2002 dataset.