Image Compression by Layered Quantum Neural Networks

S.M. Gushanskiy, Viktor Potapov, Maxim Polenov

Qigao Feng, Haoyu Zhou

Quantum Neural Network (QNN), which integrates the characteristics of Artificial Neural Network (ANN) with quantum theory, is a new study field. It takes advantages of ANN and quantum computing and has a high theoretical value and potential applications. Based on quantum neuron model with a quantum input and output quantum and artificial neural network theory, at the same time, QBP algorithm is proposed on the basis of the complex BP algorithm, the network of a 3-layer quantum BP which implements image compression and image reconstruction is built. The simulation results show that QBP can obtain the reconstructed images with better quantity compared with BP in spite of the less learning iterations.

Huifang Li, Mo Li

In this paper we combine with quantum neural networks and image compression using Quantum Gates as the basic unit of quantum computing neuron model, and establish a three layer Quantum Back Propagation Network model, then the model is used for realizing image compression and reconstruction. Since the initial weights of neural networks were slow convergence, we use Genetic Algorithm (GA) to optimize the neural network weights, and present a mechanism called clamping to improve the genetic algorithm. Finally, we combined the Genetic Algorithm with quantum neural networks to finish image compression. Through an experiment we can see the superiority of the improved algorithm.

Prateek Karkera, R Srivaramangai

Guoming Chen, Qiang Chen, Shun Long +3 more

Arun Pandian J, K. Kanchanadevi, Vadem Chandu Mohan +2 more

In this paper, a Quantum Neural Network (QNN) has been proposed using the Projected Quantum Kernel feature for an image classification task. The QCNN consists of four dense layers; the first layer collects the quantum data as an input and the fourth layer produced the classification output. Moreover, a Quantum Generative Advisory Network (QGAN) has been developed using the patching technique for enhancing the number of samples in the image dataset. The proposed QNN and QGAN are constructed using quantum filters. The MNIST handwritten digit dataset was used to train and test the QNN model performance on image classification. A binary classification dataset was created from the MNIST handwritten digit database using digits 0 and 6. The QGAN generated 221 samples on digits 0 and 6 classes. The generated samples were added to the training dataset for the QNN model. The size of the Filtered MNIST handwritten dataset was extended from 13779 to 14000 samples. There are 12,000 images are split for training and 2,000 images for testing. The principal component analysis technique was used to reduce the dimension of the data. The QNN was trained on the enhanced dataset using a GPU environment. The testing accuracy of the QNN model was 98.65 percent; it is superior to the traditional neural network.