Spatial multi-scale gradient orientation consistency for place instance and Scene category recognition

This paper presents a novel image feature called Gradient Orientation Consistency (GOC) and an image representation based on the spatial multi-scale GOC histogram for place instance and scene category recognition. Gradient orientations have proven to be a more robust descriptor of the image information than intensities, in the presence of illumination variances. The GOC feature is obtained by imposing a Local Binary Pattern (LBP) type of non-parametric operator to the gradient orientation codes, hence captures the local structures in an image. The histogram of GOC further encodes the global structure of the entire image. The multi-scale analysis and spatial pyramid techniques are also used to enrich the longer-range structure information, resulting in a comprehensive image representation. We apply this representation method for place instance and scene category recognition on five different datasets and show that it outperforms the state of the art.