Analyzing the role of visual structure in the recognition of natural image content with multi-scale SSIM

The natural continuum of visual processing from signal analysis to cognition inspires the search for extensions to current models to include some higher-level structural processing. This paper continues the authors’ study regarding the role of visual structure in cognition, where two representations of a natural image are considered: signal-based and structurally based. Structural representations preserve the overall image organization necessary to recognize the image content and discard the finer details of objects such as textures. Signal-based representations decompose and process an image as a signal in terms of its frequency, orientation, and contrast. A recognition experiment was conducted, where grayscale cartoon-renderings replace the simple line drawings previously selected as structural representations. An analysis between the two types of image representations using the three components of the multi-scale structural similarity (MSSIM) image quality metric (luminance, contrast, and structure) identifies signal structure as a crucial component in recognition.