Marcin Macias
supervisor: Przemysław Rokita
During this seminar I will present the main research area of my work, the overall idea and possible obstacles that I need to overcome.
The main goal of my Ph.D. thesis is to improve existing algorithms or propose new ones that can be used in the problem of automatic detection and comprehension of small objects on the RGB images. The real life example of such classes that I want to focus on involves labels used in the retail store that contain price information about the presented product. Those labels can be treated as anomalies on the initial image, if we compare their size with rest of the image. However, they carry a lot of valuable information that we want to collect and analyze. Additionally, this data is usually easily noticeable by a human eye, which suggests that saliency methods used for the eye tracking and gaze analysis may be integrated into this processing pipeline to improve its efficiency. My task requires usage of different image analysis, segmentation and processing techniques and each of those methods has some known drawbacks and minor imperfections that I hope to improve. After complex research of the existing literature in the first semesters of my Ph.D. studies I have moved to the implementation of the prototype of processing pipeline that will include aforementioned algorithms and my improvements. I am also in the process of collecting a real life dataset that is necessary for the future experiments and validation.