A Hybrid Approach for Counting Templates in Images
Source of Publication
ACM International Conference Proceeding Series
© 2020 ACM. In the research, hybrid algorithm for counting repeated objects in the image is proposed. Proposed algorithm consists of two parts. Template matching sub-algorithm is based on normalized cross correlation function which is widely used in image processing application. Template matching can be used to recognize and/or locate specific objects in an image. Neural network sub-algorithm is needed to filter out false positives that may occur during cross correlation function evaluation. In the last section of the paper experimental evaluation is carried out to estimate the performance of the proposed template matching algorithm for images of blood microscopy and chamomile field image. In the first case, the task is to count erythrocytes in the blood sample. In the second case, it is needed to count the flowers in the field. For all 2 datasets we got precise results that coincides with actual number of objects in image. The reason of such performance is that convolutional neural network sub-algorithm improved initial results of template-matching sub-algorithm based on correlation function.
Association for Computing Machinery
convolutional neural network, correlation function, image processing, machine learning
Ala'raj, Maher and Majdalawieh, Munir, "A Hybrid Approach for Counting Templates in Images" (2020). All Works. 142.
Indexed in Scopus