International Journal of Computer Applications |
Foundation of Computer Science (FCS), NY, USA |
Volume 184 - Number 16 |
Year of Publication: 2022 |
Authors: Amr Elmaghraby |
10.5120/ijca2022922152 |
Amr Elmaghraby . Extract Rich Information from Images and Video using Custom Vision Cognitive Services. International Journal of Computer Applications. 184, 16 ( Jun 2022), 15-28. DOI=10.5120/ijca2022922152
Computer vision is a branch of AI that allows computers and systems to extract useful information from digital photos, movies, and other visual inputs in order to address real-world visual problems. Artificial intelligence has an area called machine learning. Machine learning has a subfield called deep learning. Cognitive Services are a set of data-mining-based machine learning methods. Cognitive Machine Learning is a type of artificial intelligence that was created to address issues (AI).Deep Learning (DL)-powered computer vision technology adds real-world benefit to a variety of businesses. Deep learning is the use of neural networks containing more than one hidden layer of neurons to solve problems in domains such as computer vision which are more accurate quality inspectors than humans, make fewer mistakes, and don't mind doing tedious, repetitive duties all day. [1].Cognitive services algorithms are used in a variety of industries to help businesses and improve our daily lives. One of these domains is image classification, which uses convolutional neural networks to help humans discover key components of a picture. The purpose of this paper is to introduce the Microsoft Azure framework's Custom Vision Service. The Azure Custom Vision Service allows you to create, deploy, and develop high image identification modelsand how to make your Custom Vision Service model better. The amount, quality, and variety of labelled data you offer, as well as the entire dataset's balance, determine the quality of the classifier or object detector. A good model will have a well-balanced training dataset that is indicative of the data it will be given. The process of creating such a model is iterative, and it's normal to go through several rounds of training before getting the desired results..Convolutional neural networks, a cutting-edge technology with massive learning capacity, are used in the Custom Vision Service. Because constructing a convolutional neural network is a time-consuming activity that most engineers lack, a Custom Vision Service supplies this component for constructing a classifier. The Custom Vision service analyses photographs using a machine learning algorithm. You can use Custom Vision to create your own labels and train custom models to detect them. Each label denotes a different set of classes or objects. By submit groups of images that have and don't have the characteristics in question. The images have been labeled.at the time of submission. Then the algorithm trains this data and calculates its own accuracy by testing itself on those same images. Train the model by iterating over the entire dataset several times. On the basis of the test results, the model was evaluated. The model can be downloaded and utilized without having to be connected to the internet. Azure Cognitive Services provides a wide range of Artificial Intelligence (AI) solutions. Because the Custom Vision service is optimized for fast detecting significant differences between photographs, we can begin constructing our model with a small amount of data. We'll use 15 images in Custom Vision (the minimum required). Microsoft recommends using at least 50 different images to improve prediction accuracy (with different types of images).The suggested system can handle JPEG images, MPEG-1 bitstreams, and live video inputs. It is also possible to operate the procedures on an individual and autonomous basis. Once the training is complete, the model can be published, and you should be able to access it using the Custom Vision API..Azure Custom Vision's primary goal is to aid in the picture prediction process. The second suggested experiment, will use Java to build an integration with the Video Indexer service in order to improve it even further.