Название: System Innovation for an Artificial Intelligence Era: Applied System Innovation X
Автор: Artde D.K.T. Lam, Stephen D. Prior, Siu-Tsen Shen, Sheng-Joue Young
Издательство: CRC Press
Год: 2025
Страниц: 427
Язык: английский
Формат: pdf (true)
Размер: 16.2 MB

The book aims to provide an integrated communication platform for researchers from a wide range of topics including information technology, communication science, applied mathematics, Computer Science, advanced material science, and engineering.

The maturing Artificial Intelligence (AI) technology can produce diverse images through text prompts, useful for scene creation, conceptual modeling, and virtual illustrations. As AI art gains traction, artists increasingly employ AI image generation tools to spark or augment their creative process. This study aims to compare various AI image generation tools’ applications across fields, exploring factors like image composition and user evaluations to aid tool selection. Five representative AI image generation tools were chosen: Midjourney, Leonardo.AI, Deep AI, Stable Diffusion, and Playground AI. Through literature review, factors such as clarity, color, contrast, lighting, and composition were identified as criteria for judging image quality. Differences among these factors across tools influence generated image style, quality, and diversity.A questionnaire survey was conducted to understand image characteristics generated by different AI tools, allowing respondents to evaluate them. Subjects’ preferences and evaluations of AI-generated images may vary, influenced by their educational background and personal inclinations. A comprehensive literature review highlights differences and similarities in AI image generation tool applications across fields, delineating their respective advantages and limitations. This study’s findings offer insights into AI tool applicability across domains and inform users’ tool selection.

Deep Learning-based yoga pose assessment system: This study aims to develop a home-applicable yoga posture assessment system. Built on the MMPose Deep Learning framework, the system utilizes a general camera to instantaneously estimate the practitioner’s bodily joint coordinates in three-dimensional space. By conducting relative corrections of these three-dimensional coordinates, the system swiftly calculates changes in the bending angles of various joints. Moreover, this research employs Dynamic Time Warping (DTW) technology to compare the spatiotemporal discrepancies between user and instructor movements, thereby generating movement scores and corrective suggestions. Verified through empirical experimentation, the system developed in this study accurately provides movement scoring and offers specific recommendations for improvement. In this study, the MMPose toolkit was utilized to extract human posture information from yoga movement videos to conduct quantitative analysis and evaluate practitioner poses. MMPose is an open-source 3D human pose estimation toolkit based on the PyTorch Deep Learning framework, developed by OpenMMLab, capable of achieving high-precision 2D/3D human pose estimation.

Image cropping based on photographic composition: In photography, the precision in image composition and cropping processes can effectively identify optimal strategies for subject geometric adjustment, thereby enhancing images’ overall compositional integrity and visual appeal. In light of this, the present study introduces a novel Deep Learning system for image cropping, predicated on the principles of photographic composition, termed the Composition-based Image Cropping Network (CICNet). This system initially employs Convolutional Neural Networks (CNN) to identify key visual elements within images, subsequently utilizing Deep Learning techniques to classify photographic composition styles and generate heatmaps depicting the distribution of principal objects within the images. Following this, based on the composition style most congruent with the image, the system applies a deep cropping network to trim the original image, ensuring that principal objects are positioned in locations that align with visual aesthetic principles. Through a series of empirical validations, the system developed in this research has demonstrated its efficacy in enhancing the compositional beauty of images via automated cropping processes.








[related-news] [/related-news]