For assisting blind people to recognize their indoor objects, a new portable camera based method is introduced here. Coarse description is used here to perform the recognition task. With this technique, the presence and absence of different objects in a given query image is described. Regardless of their position within the query image it conveys the list of most likely objects present in the query image. It controls the processing time by sacrificing irrelevant information details and expands

more »

... he recognition task to multiple objects. With this technique, the perception of the blind people to his direct contextual environment can be increased. The two image multilabeling strategies which have different similarity computation is used to address the coarse description issue. First, the Euclidean distance measure is used and second one uses Gaussian process (GP) estimation which relies on semantic similarity measure. The idea is to compare the given query image with the entire set of training images that are stored offline. These offline training images are available with their associated binary descriptors. Compressive sensing provides a compact image representation to achieve fast computation capability. The two methods Euclidean Distance measure and Semantic Similarity is compared and the efficiency is studied.