A novel machine learning approach to recognize household objects
Smita Gour, Pushpa B. Patil · 2016
This work introduces a novel artificial intelligence approach to household object recognition. The approach used in this work is feature-based and it works toward recognition under a broad range of circumstances. The necessary image processing techniques are applied to recognize the objects. These techniques include removal of shadow that is segmenting the object from its shadow, extraction of shape and texture features from the object images and creation of descriptors that overcome the difficulties of affine transformations up to some extent. By this prior knowledge of descriptors, objects are categorized to their respective classes using “Back Propagation Neural Network” (BPNN). The system reached the expected result using 38 powerful combined features of shape and texture and BPNN. The system gives accuracy of 81%-92% for 10-25 different types of objects.