Image descriptions in early and mid-level vision:What kind of model is this and what kind ofmodels do we really need?

摘要

In The utility of image descriptions in the initial stages of vision: A case study of printed text,Watt and Dakin (2010) describe a model that integrates mechanisms at both early andmiddle stages of visual processing, and provide a demonstration of the application of themodel to the relational organization of printed text. In the following, we discuss anumber of the merits of this approach, but argue that it is (at this stage) highly difficult toassess the utility of this model as a plausible description of human visual processing.First, we indicate that the authors' description of the model is underspecified. Second,we question the generalizability of the model. Third, we argue that the model needs tobe directly compared to quantitative empirical data. Fourth, we argue that the modelneeds to be directly compared to alternative models.