Nicole J. J. P. Koenderink

Supporting knowledge-intensive inspection tasks with application ontologies

One of the major challenges in computer vision is to create automated systems that perform tasks with at least the same competences as human experts. In particular for automated inspection of natural objects this is not easy to achieve. The task is hampered by large in-class variations and complex 3D-morphology of the objects and subtle argumentations of experts. For example, in our horticultural case we deal with quality assessment of young tomato plants, which requires experienced specialists. We submit that automation of such a task employing an explicit model of the objects and their assessment is preferred over a black-box model obtained from modelling input-output relations only. We propose to employ ontologies for representing the geometrical shapes, object parts and quality classes associated with the explicit models. Our main contribution is the description of a method to develop a white-box computer vision application in which the needed expert knowledge is defined by: (i) decomposing the task of the inspection system into subtasks and (ii) identifying the algorithms that execute the subtasks. This method describes the interaction between the task decomposition and the needed task-specific knowledge, and studies the delicate balance between general domain knowledge and task-specific details. As a proof of principle of this methodology, we work through a horticultural case study and argue that the method leads to a robust, well-performing, and extendable computer vision system.

Requirements for a nutrition education demonstrator

[Context and Motivation] Development of innovative ICT-based applications is a complex process involving collaboration of all relevant disciplines. This complexity arises due to differences in terminology, knowledge and often also the ways of working between developers in the disciplines involved. [Question/problem] Advances in each discipline bring a rich design environment of theories, models, methods and techniques. Making a selection from these makes the development of distributed applications very challenging, often requiring a holistic approach to address the needs of the disciplines involved. This paper describes early stage requirements acquisition of a mobile nutrition education demonstrator which supports overweight persons in adopting healthier dietary behaviour. [Principal idea/results] We present a novel way to combine and use known requirements acquisition methods involving a two stage user needs analysis based on scenarios which apply a theory-based model of behavioural change and are constructed in two phases. The first phase scenarios specify an indicative description reflecting the use of the transtheoretical model of behavioural change. In the second phase, a handshake protocol adds elements of optative system-oriented descriptions to the scenarios such that the intended system can support the indicative description. [Contribution] The holistic and phased approach separates design concerns to which each of the disciplines contributes with their own expertise and domain principles. It preserves the applied domain principles in the design and it bridges gaps in terminology, knowledge and ways of working.

Expert-Based Ontology Construction: A Case-Study in Horticulture

Experts are capable of performing complex tasks in their specific field of expertise. To do this, they use a vast amount of explicit and tacit domain knowledge. For various applications it may be interesting to represent such detailed domain knowledge in a formal way. We show here the process of elicitating expert knowledge and constructing a domain ontology for a case-study in which experts assess the quality of young greenhouse plants. We have interviewed sorting experts from different plant breeders, created individual ontologies, merged these ontologies, added relevant relations from an observers point of view and checked the results in a teach-back session. We draw two main conclusions from this work. The first conclusion is that the tacit part of an experts knowledge is often explicit knowledge for another expert. The resulting merged ontology is richer than the individual ontologies. The second conclusion is that it is essential to involve an objective observer in the creation of the ontology for adding relations to the ontology that are relevant for the final purpose of the ontology, but that are part of the tacit knowledge of the experts

Unravelling the language of eating

Understanding eating behaviour requires observations of how consumers choose what to eat and of what they actually eat. Recognizing eating gestures is difficult and labour intensive. We propose an ontology of elementary eating gestures to make visual observations of eating behaviour objective and quantifiable. Moreover, this ontology facilitates automating this task. The eating gesture vocabulary is published as an ontology and is available for use.

Automatic extraction of ingredient's substitutes

Expert advice on how ingredients can be replaced in recipes is widely available on-line. However, these are general substitution rules, which do not take into account contextual factors such as culture, sensory perception, season, etc. We aim at tuning general rules to particular recipes. From an on-line food encyclopedia we extract explicit substitution rules. We also consider implicit substitution rules, derived by the categorisations in the same source. By applying Latent Dirichlet Allocation (LDA) onto a crawled dataset, we rank ingredients based on their likelihood of being interchangeable, given a recipe. The results show that our statistical approach can approximate manual judgments.