Companies all around the world leverage computer vision and deep learning algorithms to analyze images for a wide range of applications. These include tasks like image classification and anomaly or object detection, which are employed across many fields, such as healthcare and manufacturing. In certain applications, thoroughly analyzing images and their smallest details is crucial, such as identifying anomalies caused by manufacturing defects or detecting early-stage tumors by examining X-rays or MRIs. To ensure accuracy, companies often rely on high-quality images, which are typically large in size and require significant storage capacity. To develop accurate algorithms, companies need to store a huge amount of data for training and testing, which implies a great amount of storage to preserve the large-sized images. In tasks like anomaly detection, new data is often collected over time, especially when previously unseen anomalies are identified and need to be stored for further analysis, or to improve the models. This ongoing collection of data can lead to storage limitations, making it necessary to periodically expand storage space. Another side effect of dealing with large-sized images is the slowdown of the algorithm caused by the immense weight of the data. It would therefore be interesting to observe the performances of deep learning algorithms when dealing with lower-sized images. The goal of this project is to evaluate the performance of deep learning algorithms, with a focus on those used for anomaly detection, when applied to compressed versions of original images. The hypothesis is that smaller-sized images can provide significant benefits to companies by reducing storage requirements and associated costs, while also decreasing training and testing times due to the reduced data size. This project was developed in collaboration with AIVIZ SRL, where the performance of an anomaly detection algorithm in detecting production defects on pharmaceutical tablets, using compressed images, is evaluated.

Companies all around the world leverage computer vision and deep learning algorithms to analyze images for a wide range of applications. These include tasks like image classification and anomaly or object detection, which are employed across many fields, such as healthcare and manufacturing. In certain applications, thoroughly analyzing images and their smallest details is crucial, such as identifying anomalies caused by manufacturing defects or detecting early-stage tumors by examining X-rays or MRIs. To ensure accuracy, companies often rely on high-quality images, which are typically large in size and require significant storage capacity. To develop accurate algorithms, companies need to store a huge amount of data for training and testing, which implies a great amount of storage to preserve the large-sized images. In tasks like anomaly detection, new data is often collected over time, especially when previously unseen anomalies are identified and need to be stored for further analysis, or to improve the models. This ongoing collection of data can lead to storage limitations, making it necessary to periodically expand storage space. Another side effect of dealing with large-sized images is the slowdown of the algorithm caused by the immense weight of the data. It would therefore be interesting to observe the performances of deep learning algorithms when dealing with lower-sized images. The goal of this project is to evaluate the performance of deep learning algorithms, with a focus on those used for anomaly detection, when applied to compressed versions of original images. The hypothesis is that smaller-sized images can provide significant benefits to companies by reducing storage requirements and associated costs, while also decreasing training and testing times due to the reduced data size. This project was developed in collaboration with AIVIZ SRL, where the performance of an anomaly detection algorithm in detecting production defects on pharmaceutical tablets, using compressed images, is evaluated.

The Impact of Image Compression on Deep Learning Algorithms: A Study on Anomaly Detection

BARZAN, GIORGIA
2023/2024

Abstract

Companies all around the world leverage computer vision and deep learning algorithms to analyze images for a wide range of applications. These include tasks like image classification and anomaly or object detection, which are employed across many fields, such as healthcare and manufacturing. In certain applications, thoroughly analyzing images and their smallest details is crucial, such as identifying anomalies caused by manufacturing defects or detecting early-stage tumors by examining X-rays or MRIs. To ensure accuracy, companies often rely on high-quality images, which are typically large in size and require significant storage capacity. To develop accurate algorithms, companies need to store a huge amount of data for training and testing, which implies a great amount of storage to preserve the large-sized images. In tasks like anomaly detection, new data is often collected over time, especially when previously unseen anomalies are identified and need to be stored for further analysis, or to improve the models. This ongoing collection of data can lead to storage limitations, making it necessary to periodically expand storage space. Another side effect of dealing with large-sized images is the slowdown of the algorithm caused by the immense weight of the data. It would therefore be interesting to observe the performances of deep learning algorithms when dealing with lower-sized images. The goal of this project is to evaluate the performance of deep learning algorithms, with a focus on those used for anomaly detection, when applied to compressed versions of original images. The hypothesis is that smaller-sized images can provide significant benefits to companies by reducing storage requirements and associated costs, while also decreasing training and testing times due to the reduced data size. This project was developed in collaboration with AIVIZ SRL, where the performance of an anomaly detection algorithm in detecting production defects on pharmaceutical tablets, using compressed images, is evaluated.
2023
The Impact of Image Compression on Deep Learning Algorithms: A Study on Anomaly Detection
Companies all around the world leverage computer vision and deep learning algorithms to analyze images for a wide range of applications. These include tasks like image classification and anomaly or object detection, which are employed across many fields, such as healthcare and manufacturing. In certain applications, thoroughly analyzing images and their smallest details is crucial, such as identifying anomalies caused by manufacturing defects or detecting early-stage tumors by examining X-rays or MRIs. To ensure accuracy, companies often rely on high-quality images, which are typically large in size and require significant storage capacity. To develop accurate algorithms, companies need to store a huge amount of data for training and testing, which implies a great amount of storage to preserve the large-sized images. In tasks like anomaly detection, new data is often collected over time, especially when previously unseen anomalies are identified and need to be stored for further analysis, or to improve the models. This ongoing collection of data can lead to storage limitations, making it necessary to periodically expand storage space. Another side effect of dealing with large-sized images is the slowdown of the algorithm caused by the immense weight of the data. It would therefore be interesting to observe the performances of deep learning algorithms when dealing with lower-sized images. The goal of this project is to evaluate the performance of deep learning algorithms, with a focus on those used for anomaly detection, when applied to compressed versions of original images. The hypothesis is that smaller-sized images can provide significant benefits to companies by reducing storage requirements and associated costs, while also decreasing training and testing times due to the reduced data size. This project was developed in collaboration with AIVIZ SRL, where the performance of an anomaly detection algorithm in detecting production defects on pharmaceutical tablets, using compressed images, is evaluated.
Image Compression
Deep Learning
Anomaly Detection
File in questo prodotto:
File Dimensione Formato  
Barzan_Giorgia_Thesis.pdf

accesso aperto

Dimensione 12.95 MB
Formato Adobe PDF
12.95 MB Adobe PDF Visualizza/Apri

The text of this website © Università degli studi di Padova. Full Text are published under a non-exclusive license. Metadata are under a CC0 License

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/20.500.12608/80881