Welcome
This is the page of the Machine Learning project for the COMP613 course of the Technical University of Crete Master courses.
Abstract:
In this thesis a pattern recognition algorithm was designed using machine learning methods for the usage of human face detection within into-the-wild images. The problems has to be solved are not only the part of detecting a face within the image but also to classify the detections and separate them from faulty detections due to the image noise (into-the-wild images). Any shape within the image that is similar to the human face can cause this kind of faulty detections. One of the greatest problems in pattern recognition is the ratio between the ability of an algorithm to detect objects that exists and objects that do not actually exist. If the model used is a flexible generative model then usually the detection efficiency is high but the fake detections ratio is also high reducing its reliability. On the other hand when a strict model is used then the fake detections ratio is reduced but also the real detection ratio is reduced causing again the reduction of the algorithms’ reliability. The real challenge is creating a model that has high real detection efficiency and low fake detection ratio. This would offer a algorithm with a great reliability.