Randomized neural network based signature for color texture classification

Author(s)

Mesquita Sa Junior, Jarbas Joacide ^{[1, 2]} ; Backes, Andre Ricardo ^[3] ; Bruno, Odemir Martinez ^[1]

Affiliation

^[1] Univ Sao Paulo, Sao Carlos Inst Phys, POB 369, BR-13560970 Sao Carlos, SP - Brazil

^[2] Univ Fed Ceara, Dept Comp Engn, Campus Sobral, Rua Coronel Estanislau Frota 563, BR-62010560 Sobral, CE - Brazil

^[3] Univ Fed Uberlandia, Sch Comp Sci, Ave Joao Naves Avila 2121, BR-38408100 Uberlandia, MG - Brazil

• Document type: Journal article
• Source: MULTIDIMENSIONAL SYSTEMS AND SIGNAL PROCESSING; v. 30, n. 3, p. 1171-1186, JUL 2019.
• Web of Science Citations: 0

Abstract

Color texture analysis is an important subject in computer vision research. This paper presents an innovative and powerful color texture analysis method based on a randomized neural network. This approach uses the weights of the neural network as attributes for a color feature vector. Experiments were performed in three well-established benchmarks (Vistex, USPtex and Outex) and two rotated versions of these datasets (Vistex and Outex). The results were promising, surpassing the accuracies of most of the compared methods. This achievement allows us to affirm that the proposed approach is a valuable tool to be included in color texture analysis field. (AU)

• FAPESP's process: 16/18809-9 - Deep learning and complex networks applied to computer vision
• Grantee: Odemir Martinez Bruno
• Support type: Research Grants - Research Partnership for Technological Innovation - PITE
• FAPESP's process: 14/08026-1 - Artificial vision and pattern recognition applied to vegetal plasticity
• Grantee: Odemir Martinez Bruno
• Support type: Regular Research Grants