Handmade embroidery pattern recognition: a new validated database / Kudirat Oyewumi Jimoh ... [et al.]

• Main Authors: Jimoh, Kudirat Oyewumi (Author), Ọdẹ́jọbí, Ọdẹ́túnjí Àjàdí (Author), A. Fọlárànmí, Stephen (Author), Aina, Segun (Author)
• Format: Book
• Published: Universiti Teknologi MARA, 2020-06.
• Subjects: Article; PeerReviewed
• Online Access: Link Metadata
• Notes: https://ir.uitm.edu.my/id/eprint/48080/1/48080.pdf
• Keywords: Applied optics. Photonics; Optical pattern recognition

Patterns of handmade embroidery are an important part of the culture of a number of African people, particularly in Nigeria. The need to digitally document these patterns emerges in the context of its low patronage despite its quality and richness. The development of a database will assist in resuscitating the dying art of Handmade Embroidery Patterns (HEP). The patterns of handmade embroidery are also irregular and inconsistent due to the manual method, and creativity involved in its production. Developing an automatic recognition of HEP will therefore create a system where machine embroidery can be made, or automated to mimic the creativity and peculiar intricacies of traditional handmade embroidery patterns. This study developed handmade embroidery pattern database (HEPD) that can be used for many processes in the field of pattern recognition and computer vision applications. Samples of handmade embroidery patterns were collected from three different cities in South-Western, Nigeria. Pre-processing operations such as image enhancement, image noise reduction, and morphology were performed on the collected samples using image-processing toolbox in MATLAB. This work developed a validated new dataset of handmade embroidery patterns containing two categories of embroidery patterns with a total number of 315 images in the database. It evaluated the database for recognition process using cellular automata as feature extraction technique and support vector machine as its classifier. The performance metrics employed are sensitivity, specificity and accuracy. For the two classes of images considered, 72% sensitivity, specificity of 93% and accuracy of 80% were obtained for grayscale image. For the binary image, an accuracy of 72% with sensitivity of 82% and 65% specificity were obtained. The result obtained showed that the grayscale image exhibits an efficient accuracy than binary image.