From Methods to Images

This paper presents a system for automatically selecting images in an image database to be used as illustrations of an image method or analysis process. This problem is related to the teleteaching of image processing which uses already implemented libraries of algorithms. This is in the context of a teleteaching European project. We first give a Bayesian approach to this problem, by using an image basis. Then we show how to use the Haar transform for this purpose. Finally we give examples and discuss our approach. Received: June 8, 1998; revised December 17, 1998     

A Bag of Strings Representation for Image Categorization

This paper presents an architecture well suited for natural images classification or visual object recognition applications. The method proposes to integrate a spatial representation into the well known “bag of local signatures” approach. For this purpose, it combines the power of a string representation which provides an ordered view of local features with the vectorial histogram representation allowing to recognize efficiently and quickly an image by using a machine learning classifier. To reach this goal, we propose to represent an image by a set of strings of local signatures obtained by tracking the detected salient points along image edges. We propose here to conjointly use the Hölder exponents and the direction of minimal regularity of the bidimensional signal singularities to compute a signature describing precisely a region of interest centered on an interest point. As we will see, an alphabet of strings is easily obtained by using a typical self organizing map architecture. As a consequence, a “bag of strings” representation is used, providing a compact representation encoding both local signatures and spatial information. This representation is particularly well suited to train a support vector machine classifier used for the last classification step. This architecture obtains good classification rates on different well known datasets.     

An O(log n) pyramid hough transform

This paper describes a divide-and-conquer Hough transform technique for detecting a given number of straight edges or lines in an image. This technique is designed for implementation on a pyramid of processors, and requires only O(log n) computational steps for an image of size n × n.     

Images and Benford's Law

Benford's law had been proposed in the past as a way to modelize the probability distribution of the first digit in a set of natural numbers. We show in this paper that the magnitude of the gradient of an image obeys this law. We show, experimentally, that this also applies for the laplacian pyramid code. This yields to the field of entropy based coding which takes advantage of the a priori information about the probability of any symbol in the signal.     

Object count/area graphs for the evaluation of object detection and segmentation algorithms

Evaluation of object detection algorithms is a non-trivial task: a detection result is usually evaluated by comparing the bounding box of the detected object with the bounding box of the ground truth object. The commonly used precision and recall measures are computed from the overlap area of these two rectangles. However, these measures have several drawbacks: they don't give intuitive information about the proportion of the correctly detected objects and the number of false alarms, and they cannot be accumulated across multiple images without creating ambiguity in their interpretation. Furthermore, quantitative and qualitative evaluation is often mixed resulting in ambiguous measures.In this paper we propose a new approach which tackles these problems. The performance of a detection algorithm is illustrated intuitively by performance graphs which present object level precision and recall depending on constraints on detection quality. In order to compare different detection algorithms, a representative single performance value is computed from the graphs. The influence of the test database on the detection performance is illustrated by performance/generality graphs. The evaluation method can be applied to different types of object detection algorithms. It has been tested on different text detection algorithms, among which are the participants of the ICDAR 2003 text detection competition.The work presented in this article has been conceived in the framework of two industrial contracts with France Télécom in the framework of the projects ECAV I and ECAV II with respective numbers 001B575 and 0011BA66.     

Extraction and recognition of artificial text in multimedia documents

Abstract The systems currently available for contentbased image and video retrieval work without semantic knowledge, i. e. they use image processing methods to extract low level features of the data. The similarity obtained by these approaches does not always correspond to the similarity a human user would expect. A way to include more semantic knowledge into the indexing process is to use the text included in the images and video sequences. It is rich in information but easy to use, e. g. by key word based queries. In this paper we present an algorithm to localise artificial text in images and videos using a measure of accumulated gradients and morphological processing. The quality of the localised text is improved by robust multiple frame integration. A new technique for the binarisation of the text boxes based on a criterion maximizing local contrast is proposed. Finally, detection and OCR results for a commercial OCR are presented, justifying the choice of the binarisation technique.An erratum to this article can be found at     

A fast parallel algorithm for blind estimation of noise variance

A blind noise variance algorithm that recovers the variance of noise in two steps is proposed. The sample variances are computed for square cells tessellating the noise image. Several tessellations are applied with the size of the cells increasing fourfold for consecutive tessellations. The four smallest sample variance values are retained for each tessellation and combined through an outlier analysis into one estimate. The different tessellations thus yield a variance estimate sequence. The value of the noise variance is determined from this variance estimate sequence. The blind noise variance algorithm is applied to 500 noisy 256×256 images. In 98% of the cases, the relative estimation error was less than 0.2 with an average error of 0.06. Application of the algorithm to differently sized images is also discussed     

Robust clustering with applications in computer vision

A clustering algorithm based on the minimum volume ellipsoid (MVE) robust estimator is proposed. The MVE estimator identifies the least volume region containing h percent of the data points. The clustering algorithm iteratively partitions the space into clusters without prior information about their number. At each iteration, the MVE estimator is applied several times with values of h decreasing from 0.5. A cluster is hypothesized for each ellipsoid. The shapes of these clusters are compared with shapes corresponding to a known unimodal distribution by the Kolmogorov-Smirnov test. The best fitting cluster is then removed from the space, and a new iteration starts. Constrained random sampling keeps the computation low. The clustering algorithm was successfully applied to several computer vision problems formulated in the feature space paradigm: multithresholding of gray level images, analysis of the Hough space, and range image segmentation     

Hierarchical line extraction

A hierarchical line and segment extraction algorithm, based on a pyramid, is described. Initially, lines are detected in small windows using the Hough transform. The detected lines are then merged using a distance criteria thus avoiding a reaccumulation process at each level of the pyramid. The hierarchical merging process is efficiently performed on lines rather than on segments (since there are many more segments than lines). The detected lines are broken into segments, at the top of the pyramid. The proposed approach is compared to similar approaches based on hierarchical feature extraction. The authors show that their approach combines the advantages of other works and avoids their drawbacks such as quantisation effect and lack of robustness