Biologically Inspired Object Recognition using Gabor Filters

Recent advancements in the understanding of the computational processes underlying early vision have provided novel opportunities for the creation of biologically inspired vision models. Hubel and Wiesel (1968) empirically demonstrated the existence of receptive fields as a fundamental aspect of early visual processing in mammalian vision systems. Further work demonstrated that these receptive fields can be modelled as the application of linear Gabor filters on visual input data (Jones and Palmer, 1987). These advancements, and others, have led to the creation of biologically inspired feature extraction and selection techniques that emulate the early stages of mammalian visual processing (Fei-Fei et al., 2004; Jarrett et al., 2009; Pinto et al., 2008; Serre et al., 2005). In addition to these advancements in the understanding of early visual processing, there have also been a number of advancements in biologically realistic classification (LeCun et al., 2010; Woodbeck et al., 2008). There are now a large number of classifiers, the majority of which are based upon neural network architectures, that attempt to perform classification and categorization in a biologically plausible manner. In this section, I will describe the importance of biologically inspired object recognition models. I will also provide an overview of the previous work on biologically plausible object recognition and will outline some interesting questions that remain open. Lastly, I will introduce the approach I take to address some of these open questions.