Algorithms for image analysis and extraction of necessary information from them are widely applied in the scientific research and engineering solutions. Different algorithms are used to solve image recognition, segmentation, classification and other problems. Usually, image decomposition, machine learning algorithms or their combination is employed to solve a given problem. However, the application of machine learning is limited because of the big datasets required in the learning process. The construction of such dataset is complicated in the new application area. Although image analysis algorithms are well developed and their potentiality in application field is frequently discussed, the analysis of image sequence in order to identify evolution of process is challenging algorithmically. In current practice of algorithms which are used to solve semantical problems, discrete images are analyzed in order to extract objects from the image sequence and the meaning is obtained from the summarized results.
If the analysis of initial discrete image sequence is performed using image decomposition, transformation and merging algorithms, it is possible to transform the process from the discrete image sequence to continuum. The usage of transformation in continuum projected to an image or image sequence as an input in neural networks with the appropriate enables to reduce the dataset of image sequence required in the learning stage. This is an advantage in those research and engineering areas where the ability to achieve successful results is limited by small datasets because it is not possible to collect more data. In addition, if large datasets are used in the learning process, better results can be obtained compared to the traditional algorithms where the evolution of the process is evaluated from the generalized results of discrete image sequence analysis.
Project funding:
KTU R&D&I Fund
Period of project implementation: 2019-04-01 - 2019-12-31
