http://dspace.epoka.edu.al/handle/1/1662 2026-04-20T09:34:51Z 2026-04-20T09:34:51Z Tare, Aleks http://dspace.epoka.edu.al/handle/1/2469 2025-01-24T12:23:21Z 2020-07-13T00:00:00Z

MICROSCOPIC IMAGE CELL COUNTING USING CONVOLUTIONAL NEURAL NETWORKS Tare, Aleks As the field of automation is moving forward at ever-faster rates, cell counting and classification is an omnipresent yet repetitive task that would benefit greatly from this field. The counting of contiguous cells in a specific area could provide crucial contribution to work done in clinical trials. Cell counting, sadly, is most often conducted manually by humans and can be time and resource consuming. Due to cells touching each other, a non-uniform background, shape and size variations of cells, and different techniques of image acquisition, the task becomes even more difficult. In this paper we describe a convolutional neural network approach, using a Faster-RCNN architecture later also combined with a U-Net neural network, for cell counting and possibly segmentation in a raw microscopic picture.

2020-07-13T00:00:00Z Lisi, Aldo http://dspace.epoka.edu.al/handle/1/2468 2025-01-24T12:21:41Z 2020-07-21T00:00:00Z

FRACTAL IMAGE COMPRESSION USING NEURAL NETWORKS Lisi, Aldo Image compression is an interesting field in image analysis that has been around for quite a long time now. This field simply aims to reduce the image size and to maintain a good level of their reconstructed image. In the image compression field there have been a lot of techniques about reducing the image size and reconstructing it as close as the original one. Some of those techniques are quite old now and they are still being used. The main problems in image compression field and those techniques is the encoding and decoding time. Meanwhile the compression ratio is quite impressive even for RGB images. One of those techniques is compressing images using fractals. Here we are going to see fractal image compression based on techniques related to calculating partial distance between domain and range blocks and neural network for feature selection.

2020-07-21T00:00:00Z Panci, Ina http://dspace.epoka.edu.al/handle/1/2467 2025-01-24T12:18:50Z 2020-07-13T00:00:00Z

CELL DETECTION USING DEEP LEARNING AND HAND-CRAFTED FEATURES Panci, Ina With the Artificial Intelligence becoming more and more powerful in time, there is a growing interest about scientists focusing in providing answers in the field of medical image analysis. The importance of using AI in the medical field is of great significance for many reasons. Robust algorithms are able to help physicians in processing large amounts of data, assisting in providing diagnosis and mitigating human error while reducing time and cost. This is especially applicable in the case of cell image analysis. Considering the complexity of cells as entities, which derives from their shape and form, it can be quite challenging to detect them in different settings that they are placed. An automated process on cell detection will be very beneficiary for anatomopathologists. In this work, there is cell detection performed, using the state-of-the-art deep learning object detection model, Faster R-CNN and trained on hand-crafted features such as Local Binary Patterns. The dataset used portrayed several challenges that most microscopy images hold, which required histogram and template matching for preprocessing. With cell detection performed, we also count the cells in an image by nuclei localization. There were several experiments conducted that achieve up to 56% mAP.

2020-07-13T00:00:00Z Koçi, Jola http://dspace.epoka.edu.al/handle/1/2466 2025-01-24T12:16:46Z 2020-07-23T00:00:00Z

A REVIEW THREAT OBJECT DETECTION IN X -RAY IMAGES USING SSD, R-FCN AND FASTER R -CNN Koçi, Jola Baggage inspection for threat objects using X-ray images is a priority task that is in charge of making the risk of crime and terrorist attacks more reducible. Nowadays, the checking of baggage is based on a semi-automated system that consists of both human and also image detection. The main purpose of this thesis is to make the automatisation system more reliable. This task is mainly dependent on object detection models and algorithms. However, another part of this problem is mainly prone to the lack of the data. Furthermore, obtaining an X-ray image dataset with many types of threat objects is quite difficult. This is why this thesis is composed of two main parts: data stimulation and object detecting approaches. On the first part, due to the lack of data, an older dataset containing only four classes of threat objects are used as a base for the new objects to be stimulated into. In total, the newly simulated dataset contains seven types of threat objects consisting of: handguns, razor-blades, knife, shuriken, battery, wires and mortar. After generating the new data, they are processed and augmented by applying random types of rotations, flippings, zoomings etc. Once the images are processed, they are passed into transfer learning. Transfer learning consists of using predefined models for training. The models that are taken into consideration are: Single Shot Detector, Regions Fully Convolutional Network (R- FCN) and Faster R-CNN. These models are used by applying different techniques of feature extraction, such as: Inception-v2, MobileNet-v2 and ResNet101. Combining the object detection models and object detection architectures in total the images are trained and tested in five different approaches. In conclusion, the best detection was achieved by the combination of Faster-RCNN detection model and ResNet101 feature extractor by 87.58% ± 0.75 accuracy.

2020-07-23T00:00:00Z