a Computational Approach to Estimation of Crowding in Natural Images
Lauri Ahonen · viXra · 2018
Crowding is a phenomenon where the identification of objects in peripheral vision is deteriorated by the presence of nearby targets. Crowding therefore reduces the extent of visual span, i.e. information intake during a single eye fixation. It is, thus, a limiting factor of many everyday tasks, such as reading. The phenomenon is due to wide area feature integration in the higher levels of visual processing. Despite the critical role of the phenomenon, complex natural images have so far not been used in the research of crowding. The purpose of the present study was to determine how the crowding effect affects object recognition in complex natural images, and whether the magnitude of the crowding could be modelled using methods introduced below. The actual magnitude of the crowding effect was determined experimentally by measuring contrast thresholds for letter targets of different sizes on various natural image backgrounds. The results of the experiments were analyzed to evaluate the developed methods. The methods are based on image statistics and clutter modelling. Clutter models assess the complexity in the image. The image statistics and the clutter models were combined with basic knowledge of the crowding effect. In addition, an early visual system model was incorporated to assess the role of the visual acuity across the visual field. The developed models predicted the induced crowding effect in an arbitrary natural image. The model of the visual system contributed to the results, as well. The differences between the methods for assessing the image properties were, however, negligible. Contrast energy, the simplest measure, can be regarded as the most efficient. Natural images can cause very strong crowding effects. The conclusion is that predicting quantitative dimensions of the crowding effect in an arbitrary image is viable. However further research of the subject is necessary for developing the models. Computational assessment of the crowding effect potentially can be applied to e.g. user interface design, assessing information visualization techniques, and the development of augmented reality applications.