Image Classiﬁcation and retrieval of image from a large database has a great relevance in the present Scenario. A lot of work for an eﬃcient method of image retrieval from large database has been made in the recent surveys. Here we propose a mathematical model based on CBIR system that uses the deep neural architecture for classiﬁcation where the inputs are fuzzy grassland image features. Grassland image features varies according to the varieties of grassland images available through satellite images and hence its classiﬁcation is a complex process. This paper proposes a new method for classiﬁcation in which the inputs to the Neural Network are fuzziﬁed and transformed in such a way that it clusters around a pivot vector there by making the classiﬁcation task less complicated. This classiﬁcation procedure is established theoretically by developing a mathematical model based on Neural Network approximation with fuzzy inputs. This model brings a transformation from the input image feature space to the output approximation space through the composition of mapping between the hidden transformation spaces that helps to strengthen the function approximation to the desired output. The Graphical representation on Fig(i) throws an insight into the mathematical theory of a CBIR system which uniﬁes the advantages of deep neural architecture and fuzzy approximators. The mathematical concepts such as open balls, metric, limits, continuity etc are incorporated to establish the necessary and suﬃcient condition in the fuzzy based neural system for better and clear image retrieval.
Digital Object Identifier (DOI)
Gopalan, Sasi; Pinto, Linu; C., Sheela.; and Kumar M. N., Arun
"Function Approximation with Deep Neural Network for Image Classiﬁcation in Fuzzy Domain,"
Applied Mathematics & Information Sciences: Vol. 11
, Article 19.
Available at: https://dc.naturalspublishing.com/amis/vol11/iss6/19