Abstract:
Quantum computing has brought new horizons in solving problems that are challenging for
traditional techniques and approaches. A promising application of quantum computing in the
f
ield of digital image processing is the use of Quantum Convolutional Neural Networks (QC
NNs) for image classification. This thesis aims to enhance the efficacy of image classification
by combining quantum and classical computing. The study focuses on the application of the
QCNN, a quantum analog of the classical convolutional layer, to improve the functionality of
conventional CNNs. This approach seeks to exploit the phenomena of quantum entanglement
and superposition during the feature extraction process of traditional CNNs. The implementa
tion of QCNNsinvolves creating quantum qubits and quantum circuits using the Qiskit platform
on IBM’s quantum computing system. The QCNN’s performance is evaluated on the MNIST
database, a standard benchmark for image classifiers by using Pennylane.The initial phase of
our research involved utilizing a limited dataset, which previously achieved a benchmark accu
racy of 79 % over the last decade. Comparative analysis with classical CNNs is also performed
to assess the efficacy of the quantum computing technique. The results indicate that the QCNNs
can achieve competitive accuracy with less computational cost, potentially outperforming clas
sical models under certain conditions. Building upon this foundation, we expanded our study
to incorporate a full dataset, significantly improving model performance. By employing angle
encoding with four qubits, we enhanced the model’s accuracy and demonstrated the potential of
quantum computing to revolutionize image processing. By offering extensive insights into QC
NNs,angle encoding and their implications for image classification tasks,represent a promising
direction and pave the way for practical quantum machine learning application.
