Frédéric Rayar

An Efficient Coarse-to-Fine Indexing Technique for Fast Text Retrieval in Historical Documents

In this paper, we present a fast text retrieval system to index and browse degraded historical documents. The indexing and retrieval strategy is designed in a two level, coarse-to-fine approach, to increase the speed of the retrieval process. During the indexing step, the text parts in the images are encoded into sequences of primitives, obtained from two different codebooks: a coarse one corresponding to connected components and a fine one corresponding to glyph primitives. A glyph consists of a single character or a part of a character according to the shape complexity. During the querying step, the coarse and the fine signature are generated from the query image using both codebooks. Then, a bi-level approximate string matching algorithm is applied to find similar words, using coarse approach first, and then the fine approach if necessary, by exploiting predetermined hypothetical locations. An experimental evaluation on datasets of real life document images, gathered from historical books of different scripts, demonstrated the speed improvement and good accuracy in presence of degradation.

CNN Training with Graph-Based Sample Preselection: Application to Handwritten Character Recognition

In this paper, we present a study on sample preselection in large training data set for CNN-based classification. To do so, we structure the input data set in a network representation, namely the Relative Neighbourhood Graph, and then extract some vectors of interest. The proposed preselection method is evaluated in the context of handwritten character recognition, by using two data sets, up to several hundred thousands of images. It is shown that the graph-based preselection can reduce the training data set without degrading the recognition accuracy of a non pretrained CNN shallow model.

Visual Analysis System for Features and Distances Qualitative Assessment: Application to Word Image Matching

In this paper, a visual analysis system to qualitatively assess the features and distance functions that are used for calculating dissimilarity between two word images is presented. Computation of dissimilarity between two images is the prerequisite for image matching, indexing and retrieval problems. First, the features are extracted from the word images and a distance between each image to others is computed and represented in a matrix form. Then, based on this distance matrix, a proximity graph is built to structure the set of word images and highlight their topology. The proposed visual analysis system is a web based platform that allows visualisation and interactions on the obtained graph. This interactive visualisation tool inherently helps users to quickly analyse and understand the relevance and robustness of selected features and corresponding distance function in a unsupervised way, i.e. without any ground truth. Experiments are performed on a handwritten dataset of segmented words. Three types of features and four distance functions are considered to describe and compare the word images. Theses material are leveraged to evaluate the relevance of the built graph, and the usefulness of the platform.

On Fast Sample Preselection for Speeding up Convolutional Neural Network Training

We propose a fast hybrid statistical and graph-based sample preselection method for speeding up CNN training process. To do so, we process each class separately: some candidates are first extracted based on their distances to the class mean. Then, we structure all the candidates in a graph representation and use it to extract the final set of preselected samples. The proposed method is evaluated and discussed based on an image classification task, on three data sets that contain up to several hundred thousands of images.

An Image-Based Representation for Graph Classification.

This paper proposes to study the relevance of image representations to perform graph classification. To do so, the adjacency matrix of a given graph is reordered using several matrix reordering algorithms. The resulting matrix is then converted into an image thumbnail, that is used to represent the graph. Experimentation on several chemical graph data sets and an image data set show that the proposed graph representation performs as well as the state-of-the-art methods.

APoD eXplorer: Recommendation System and Interactive Exploration of a Dynamic Image Collection

The amount of captured images has increased exponentially these last years. Online available image collections are becoming common thanks to social networks or institutes digitization programs. The context of our work falls into the need to explore such image collections. The literature paradigms are leveraged to meet three constraints: (i) handling medium to large image collections, (ii) handling dynamic image collections and (iii) providing interactive visualisations. In this paper, we describe how our work has been used to build a recommendation system and an interactive exploration platform for dynamic image collection. To illustrate the relevance of such tools, we present APoD eXplorer, an online available platform that enhances the exploration of the NASA Astronomy Picture of the Day image collection. The platform is available at http://frederic.rayar.free.fr/apod/.