ANW Rev. Neurmci. 19%. 19:109-39 Copyrighr 0 I996 by Annual Reviews Inc. AU righrs reserved INFEROTEMPORAL CORTEX AND OBJECT VISION Keiji Tanaka The Institute of Physical and Chemical Research (RIKEN), 2-1 Hirosawa, Wako-shi, Saitama, 35141, Japan KEY WORDS: macaque monkey, extrastriate visual cortex, object vision, optical imaging, population coding ABSTRACT Cells in area TE of the inferotemporal cortex of the monkey brain selectively respond to various moderately complex object features, and those that cluster in a columnar region that runs perpendicular to the cortical surface respond to similar features. Although cells within a column respond to similar features, their selectivity is not necessarily identical. The data of optical imaging in 'E have suggested that the borders between neighboring columns are not discrete; a continuous mapping of complex feature space within a larger region contains several partially overlapped columns. This continuous mapping may be used for various computations, such as production of the image of the object at different viewing angles, illumination conditions. and articulation poses. Introduction Recognizing objects by their visual images is a key function of the primate brain. This recognition is not a template matching between the input image and stored images but a flexible process in which considerable change in imagesaue to different illumination, viewing angle, and articulation of the objectdm be tolerated. In addition, our visual system can deal with images of novel objects, based on previous visual experience of similar objects. Gen- eralization may be an intrinsic property of the primate visual system. In this article, I discuss the neural organization essential for these flexible aspects of visual object recognition in the anterior part of the inferotemporal cortex. The inferotemporal cortex (IT) of the monkey brain has been divided into subregions in several different manners. Our own division into posterior IT and anterior IT, based on the size of the receptive fields and the properties of responses (Tanaka et a1 1991, Kobatake & Tanaka 1994). roughly corresponds to the previous cytoarchitectural division into TEO and TE (Iwai & Mishkin 1967; von Bonin & Bailey 1947,1950): Posterior ITcorresponds to EO, and 109 0147-006)(/96/0301-0109$08.00 Annual Reviewswww.annualreviews.org/aronlineAnnu. Rev. Neurosci. 1996.19:109-139. Downloaded from arjournals.annualreviews.orgby Princeton University Library on 01/30/08. For personal use only.110 TANAKA anterior IT to TE. I use TEO and TE in this article because they are more TE receives visual information from the primary visual cortex (Vl) through a serial pathway, which is called the ventral visual pathway (Vl-V2-V4- TEO-TE). Although there are also jumping projections, such as that from V2 to TEO (Nakamura et a1 1993) and that from V4 to the posterior part of TE (Saleem et a1 1992). the step-by-step projections are more numerous. The IT projects to various brain sites outside the visual cortex, including the perirhinal cortex (areas 35 and 36), the prefrontal cortex, the amygdala, and the striatum of the basal ganglia. The projections to these targets are more numerous from TE, especially from the anterior part of TE, than from the areas at earlier stages (Iwai & Yukie 1987, Ungerleider et a1 1989, Saleem et al 1993% Cheng et al 1993, Suzuki & Amaral 1994). Therefore, there is a sequential cortical pathway from V1 to TE, and outputs from the pathway originate mainly in TE. Monkeys that have had their TE bilaterally ablated showed severe but selective deficits in learning tasks that required the visual recognition of objects (Gross 1973, Dean 1976). These behavioral results, together with the above- described important anatomical position of TE, suggest that TE is the site of neural organization essential for the flexible properties of visual object recog- nition. In this review, our own data are emphasized, and the citation of other references is selective. This selection is not based only on the value of the studies but also on their relevance to the subject. The readers should read other reviews to get an overview of studies in the IT, e.g. Rolls (1991), Miyashita (1993), Gross (1994), and Desimone et al (1994). In particular, mechanisms of short-term memory of object images are not discussed in this article. I first summarize the data from unit-recording experiments to show that cells in TE respond to moderately complex object features and that those that cluster in a columnar region respond to similar features. I then consider the process by which the selectivity is formed in the afferent pathways to TE. I introduce the data of optical imaging of TE in order to discuss the function of the TE columns. Finally. I consider how the concept of the object emerges in the brain. The selections of our recordings that are introduced in this article were all conducted in anesthetized preparation, and they were from the lateral part of TE, lateral to the anterior middle temporal sulcus (AMTS). This part is often referred to as TEd (dorsal part of TE). popular. Stimulus Selectivity of Cells in TE One obstacle in the study of neuronal mechanisms of object vision has been the difficulty in determining the stimulus selectivity of individual cells. There Annual Reviewswww.annualreviews.org/aronlineAnnu. Rev. Neurosci. 1996.19:109-139. Downloaded from arjournals.annualreviews.orgby Princeton University Library on 01/30/08. For personal use only.INFEROTEMPORAL CORTEX 11 1 is a great variety of object features in the natural world, and we do not know how the brain scales down the dimension of this variety. Single-unit recordings from TE were initiated by Gross and his colleagues (Gross et al1969,1972). They found that cells in TE had large receptive fields, most of which included the fovea, and that some cells responded specifically to a brush-like shape with many protrusions or to the silhouette of a hand. They extended the study of the stimulus selectivity by using two different methods: a constructive method and a reductive one. In the constructive meth- od, they used Fourier descriptors that were defined by the number (frequency) and amplitude of periodic protrusions from a circle. Any contour shape can be reconstructed by linearly combining elementary Fourier descriptors of sin- gle