Jean-Bernard Martens

The Hermite transform-theory

The author introduces a scheme for the local processing of visual information, called the Hermite transform. The problem is addressed from the point of view of image coding, and therefore the scheme is presented as an analysis/resynthesis system. The objectives of the present work, however, are not restricted to coding. The analysis part is designed so that it can also serve applications in the area of computer vision. Indeed, derivatives of Gaussians, which have found widespread application in feature detection over the past few years, play a central role in the Hermite analysis. It is also argued that the proposed processing scheme is in close agreement with current insight into the image processing that is carried out by the human visual system. In particular, it is demonstrated that the Hermite transform is in better agreement with human visual modeling than Gabor expansions. >

The Hermite transform-applications

It is demonstrated how the Hermite transform can be used for image coding and analysis. Hierarchical coding structures based on increasingly specified basic patterns, i.e. general 2-D patterns, general 1-D patterns, and specific 1-D patterns such as edges and corners, are presented. In the image coding application, the relation with existing pyramid coders is described. A new coding scheme, based on local one-dimensional image approximations, is introduced. In the image analysis application, the relation between the Hermite transform and existing line/edge detection schemes is described. It is shown that, by concentrating on more specific patterns, the coding efficiency can be increased since fewer coefficients have to be coded. Meanwhile, sufficient descriptive power can be maintained for approximating the most interesting features in natural images. >

Judging correlation from scatterplots and parallel coordinate plots

Scatterplots and parallel coordinate plots (PCPs) that can both be used to assess correlation visually. In this paper, we compare these two visualization methods in a controlled user experiment. More specifically, 25 participants were asked to report observed correlation as a function of the sample correlation under varying conditions of visualization method, sample size and observation time. A statistical model is proposed to describe the correlation judgment process. The accuracy and the bias in the judgments in different conditions are established by interpreting the parameters in this model. A discriminability index is proposed to characterize the performance accuracy in each experimental condition. Moreover, a statistical test is applied to derive whether or not the human sensation scale differs from a theoretically optimal (that is, unbiased) judgment scale. Based on these analyses, we conclude that users can reliably distinguish twice as many different correlation levels when using scatterplots as when using PCPs. We also find that there is a bias towards reporting negative correlations when using PCPs. Therefore, we conclude that scatterplots are more effective than parallel plots in supporting visual correlation analysis.

Design research & tangible interaction

The research on Tangible Interaction (TI) has been inspired by many different disciplines, including psychology, sociology, engineering and human-computer interaction (HCI). Now that the field is getting more mature, in the sense that basic technologies and interaction paradigms have been explored, we observe a growing potential for a more design-oriented research approach. We suggest that there are several arguments for this proposed broadening of the TI-perspective: 1) the need for designing products within contexts-of-use that are much more challenging and diverse than the task-oriented desktop (or tabletop) systems that mostly inspire us today, 2) the interest to also design TI starting from existing physical activities instead of only as add-ons to digital applications, 3) the need for iterative design and evaluation of prototypes in order to develop applications that are grounded within daily practice over prolonged periods of time, and 4) the need to extend ease-of-use to more hedonic aspects of interaction such as fun and engagement

A computer support tool for the early stages of architectural design

Even in the current computer age, there are still many important application areas, such as early architectural design, where traditional tools like sketching on paper continue to be preferred by many professionals over computer-based tools. There is a growing awareness that there are often very good grounds for this preference. Hence, instead of trying to replace such traditional ways of working, it is now often considered more opportune to try and preserve the strengths of these traditional ways of working, while at the same time improving them by providing access to new media. This is one of the main objectives of the augmented reality approach that we adopt here. In this paper, we specifically discuss the realization of a tool for early architectural design on an existing augmented reality system, called the Visual Interaction Platform. We describe the development process, the resulting tool and its performance for elementary tasks such as positioning and overdrawing. We also identify directions for future research and applications.

Perceptual characterization of images degraded by blur and noise: experiments

The notion of a perceptual space is useful for characterizing images in terms of their perceptual image quality. It is shown that images degraded by blur and noise can be assigned positions in a multidimensional perceptual space and that quality and its underlying attributes sharpness and noisiness can be associated with directions (or vectors) in this space. The perceptual space is constructed using multidimensional scaling (MDS) techniques. Two different MDS approaches are used, one making use of the perceived dissimilarity between the images and another making use of the scaled perceptual attributes sharpness and noisiness. The two alternative approaches lead to stimulus configurations that can be approximately related by a linear transformation, and the results from both MDS approaches can be combined to produce one perceptual space for each scene. A two-dimensional perceptual space adequately represents the processed images for all three scenes that are used, and the perceptual spaces obtained for all scenes are very similar. The directions of the attribute vectors in the perceptual space indicate that unsharpness and noisiness are approximately orthogonal attributes, which implies that there is little interaction between these attributes. The impairment vector, whose direction is opposite to that of the quality vectors, lies between the unsharpness and noisiness vectors at an angle of approximately 30 degrees between the unsharpness and impairment vectors.

Image representation and compression with steered Hermite transforms

In this paper we propose that the Hermite transform can bridge the gap between conventional transform coding techniques and second-generation coding techniques that make use of explicit descriptions of perceptually important image structures (e.g., edge contours and lines). This claim is motivated by two important characteristics of the Hermite transform. First, it shows good image analysis properties in detecting and extracting (local) image primitives such as orientation and position. Second, extracted image features can be used to steer the Hermite transform into a form with high-energy compaction properties. To demonstrate the efficiency of the Hermite transform in image compression, we present an image compression scheme based on a Hermite transform that adapts to local image orientations. Comparisons with other compression techniques such as JPEG show that the proposed scheme performs extremely well at high compression ratios, not only in terms of peak-signal-to-noise ratio but also in terms of perceptual image quality.

Perceptual characterization of images degraded by blur and noise: model.

Reliable and economic methods for assessing image quality are essential for designing better imaging systems. Although reliable psychophysical methods are available for assessing perceptual image quality with the help of human subjects, the cost of performing such experiments prevents their use for evaluating large amounts of image material. This has led to an increasing demand for objective methods for estimating image quality. The perceived quality of an image is usually determined by several underlying perceptual attributes such as sharpness and noisiness. In the accompanying paper [J. Opt. Soc. Am. A 13, 1166-1177 (1996)] it is demonstrated that the relationships between images on the one hand and judgments on attributes and overall quality by subjects on the other hand can be characterized in a multidimensional perceptual space. In this perceptual space the images are represented by points, and the strengths of their perceptual attributes are modeled by the projections of these image positions onto the attribute axes. In analogy with the perceptual space we will introduce a psychometric space in which the positions of the images are determined by objective measures on the images. In the case of images degraded by blur and noise the stimulus coordinates are functions of the estimated spread of the blurring kernel and the estimated standard deviation of the noise, respectively. According to the model presented in this paper, the perceptual attributes of images can be estimated in three steps. In the first step the physical parameters (blur spread and noise standard deviation) are estimated from the images. In the second step these estimates are used to position the images in psychometric space. In the third step the attribute strengths are derived by projecting the alter image positions onto the attribute axes. We show that the attributes and the quality thus estimated correlate well with the perceived attributes and quality.

Interaction techniques for navigation through and manipulation of 2D and 3D data

In this article we present a working prototype incorporating some new interaction techniques for the navigation through and the manipulation of both 3D and 2D data. The prototype aims at professional applications like architectural design, surgical planning and geological exploration. Its design was influenced by the analysis of user requirements and by the requirement for a natural interface. The prototype permits the user to navigate through 3D and 2D data in order to explore the internal structure. 3D navigation is accomplished by means of a natural selection of dynamically updated cross-sections. The user can navigate through these selected cross-sections of 2D data and can perform manipulations with them. As a proof of concept we focus on two disciplines, viz. architectural design and surgical planning. Our usability evaluation of the interaction styles demonstrates that they are perceived as useful and enjoyable additions to existing techniques in these disciplines.

Subjective quality assessment of compressed images

With limited additional effort, one can obtain valuable information on the subjective performance of image compression schemes. The choice of an appropriate experimental technique, however, is not always trivial, since both the subjects task and the experimental method itself should depend on the properties of the stimulus set (quality range, kind of artifacts). The goal of this paper is to demonstrate how to use a number of numerical scaling techniques in assessing perceived image quality. We show how to analyze data obtained in numerical category scaling experiments and how to set up such experiments. The results of several subjective experiments illustrate that numerical category scaling techniques provide an efficient means not only for obtaining compression ratio versus quality curves that characterize coder performance over a broad range of compression ratios, but also for assessing perceived image quality in a much smaller range (e.g., close to threshold level). However, the nature of the artifacts introduced by different coders can cause problems when evaluations are carried out using direct numerical category scaling. The latter is demonstrated by comparing the results of a direct scaling method and a scaling technique in which subjects have to determine quality differences between all possible combinations of coded images.