Web content is diverse and is regarded as the primary source of accessible information that can be accessed through reference links. Web facial images are one type of web content that relates to important web pages and is considered important information for individuals. This chapter proposes face recognition as a service architecture that is based on real-world images from the web. The proposed service is implemented as a service for other third parties via cloud computing; additionally, its architecture is built via cloud using virtual machines that can be expanded based on resource demands. Web crawlers crawl web pages and retrieve images for elastic cloud storage. The collected images are then used to remove human faces and prepare the face images for identification and identifying the matched face of the set through successive phases. This chapter used PCA for features extraction and KNN for identification. Experiments show that increasing the number of crawler instances improves crawling speed and improves face recognition accuracy by preferring Euclidean over other metrics.
Top1. Introduction
In recent decades, the World Wide Web has emerged as the primary source of information; moreover, it is distinguished by a wide range of services. These resources are defined by URLs, which link to other resources. As a result, systems must monitor these URLs and their contents in order to index them and make them searchable. These are web crawling engines that are designed for fetching and collect web contents. The web crawling engines are designed for focusing on a particular form of content in the web pages.
The web page contents are also diverse, as they can be paragraphs, pictures, diagrams, or tabular content. Despite the fact that older engines were not based on the using of photos on web pages to search and instead focused on the paragraphs sections of every page (Rodriguez, 2014, pp. 1274-1285). As a consequence, search engines have recently introduced new capabilities that are based on this content, such as images based search engine. The photos' content might be for a number of items. As a result, image processing is used to look for photos on the internet (Jing, 2008, pp. 307-316). The image analysis methodology used varies depending on the picture categories.
Face recognition in the World Wide Web is a challenging task. As a result, there are numerous facial recognition algorithms require ordered set of training and provide varying planes of accuracy. Since web images are disorganized and unclassified, images downloaded from WWW must be arranged and classified. As a result, search for a humanoid facial on WWW necessitates two steps: the first is downloading pictures from WWW, and the next is enquiring the picture. Furthermore, since billions of pictures existed on the internet, facial recognition jobs require extendibility, precision, and rapidity (Ortiz, 2014, pp. 153-170).
Face recognition is a visual recognition problem that consists of four main modules as shown in fig 1; face detection, preparation, feature extraction, and identification. Face detection is responsible of check if the input image contains any faces or not, then extract the faces if exist. The face preparation phase preprocesses the extracted faces by cropping and resizing to expose and focus on a proper image for the next phase. Feature extraction is responsible of representing the face image in a features list in vector manner that can be quantized and compared with other. The extracted features vector is passed to identification phase to elect the nearest and the most matched face from a set of prepared faces (ElAraby, 2021, pp. 11723–11738).
The PCA algorithm is regarded as one of the most common methods for face recognition that is built on data decrease. Eigen faces are a statistical method that denotes the faces as a linear pattern of weighted eigenvectors (Poon, 2011, pp. 245-259), (Bahurupi, 2012, pp. 91-94). The eigenvectors are derived from a training set's covariance matrix. Eigenvectors introduced a new face space in which images are depicted.
In addition, PCA is a valid method for feature selection and can select the most important variables that have most variation. Moreover, it is useful to speed up the computation and to improve the accuracy of classification model by reducing the dimensionality of the data and when there is a high dimensionality with high correlated variable (Anter, 2015, pp. 89-94), (Anter, 2015, pp. 89-94).