SUFID: Sliced and Unsliced Fruits Images Dataset

• Published in: Advances in Visual Informatics Vol. 11870; pp. 237 - 244
• Main Authors: Turayki, Latifah Abdullah Bin, Abubacker, Nirase Fathima
• Format: Book Chapter
• Language: English
• Published: Switzerland Springer International Publishing AG 2019; Springer International Publishing
• Series: Lecture Notes in Computer Science
• Subjects: Balanced fruits dataset; Multiclass fruits; Sliced fruits; Unsliced fruits
• ISBN: 9783030340315; 3030340317
• ISSN: 0302-9743; 1611-3349
• DOI: 10.1007/978-3-030-34032-2_22

Given the recent surge in online images of fruit, ever more sophisticated models and algorithms are required to organize, index, retrieve, and interact such data. However, the relative immaturity of the processes in current use is holding back development in the field. The current paper explains how images of ten classes of fruit namely apples, bananas, kiwis, lemons, oranges, pears, pineapples, coconuts, mangos and watermelons, presented both sliced and unsliced, were sourced from the Fruits360, FIDS30, and ImageNet datasets to create a single database, “SUFID”, containing 7,500 high-res images. Pre-processing was based on using pixel color distribution to determine whether each image was corrupt, or would fit the database. The paper describes the unique opportunities, principally within computer vision, presented by SUFID’s hierarchical structure, accuracy, diversity, and scale, all of which can be of use to food researchers. As the dataset and benchmarks are believed to be of benefit to all researchers in the field, they are offered gratis to support future study.