Although neurons in columns of tissue in visual cortex of adult carnivores and primates share similar preferences for the orientation of a visual stimulus, they can nevertheless have sparse and temporally uncorrelated firing visual response properties. This effect is supported by the observation that long-range excitatory connections between cortical neurons are functionally specific, because they interconnect columns with similar orientation preference, and local short-range ones are unspecific and sparse. Coupled with strong local inhibition, this network architecture is a good recipe for local competition arranged within a cortical column. In this paper we propose a model architecture that is consistent with these experimental and anatomical findings, and which explains the emergence of orientation-tuned surround suppression. We explore the effect of local columnar competition, coupled with local and long-range functional specificity, as substrates for integration of responses from the visual surround in columnar visual cortex. In addition, we show how presentation of simulated full-field complex visual stimuli, designed to approximate visual scenes, leads to reduced correlation of local excitatory responses and increased excitatory response selectivity (lifetime sparseness). These effects occurred simultaneously with increased inhibitory activity and decreased inhibitory sparseness, consistent with recordings of excitatory and inhibitory neurons in cortex. In our networks competition, reduced correlation and increased sparseness depended on both local and long-range specific excitatory connectivity. The mechanism of local competition implemented in our model explains several aspects of integration of information from the visual surround, and is consistent with experimentally measured spatial profiles of cortical excitation and inhibition.