Image processing to facilitate image identification by partially-sighted

Abstract A previous study from our laboratory demonstrates that central visual field loss produces interference in identifying objects embedded in textured backgrounds. These results motivated us to investigate segmentation algorithms that would permit separate treatment of figure and ground, in order to enhance selectively the identifiability of objects. The segmentation algorithms that we used are based on coding the image as a “steerable pyramid,” e.g., of Portilla and Simoncelli or with Liu's spectral histogram method. These methods allow us to define a profile of different areas of an image based upon textural properties related to orientation and spatial frequency differences, which then serve as criteria for figure/ground segmentation. In order to verify the previous results, we evaluated the minimal bandwidth for identification with pre-segmented and unsegmented images in AMD patients and in foveal vision of normal controls. A real-time image-processing tool could be a valuable visual aid for partially-sighted observers.