Quantum Inspired Image Classification: A Hybrid SVM Framework

We propose a Hybrid Quantum Classical Algorithm (HQCA) for image classification using a Quantum Kernel which is efficient to produce than a traditional kernel for performing the kernel trick required in Support Vector Classification. Since classical data cannot be directly provided to a quantum circuit, amplitude encoding is performed on the image matrix. Using the ZZFeatureMap of Qiskit framework we store the extracted feature of the images. ZZFeatureMap is chosen because of its promising behaviour towards providing quantum advantage. Using this feature map we create a quantum kernel which can be fine-tuned to a ground level for a particular data set by the programmers. We use FidelityQuantumKernel class from qiskit library to create the kernel. To reduce the size of the dataset required to train the model we use a image augmentation layer for data expansion. Python library called Augmentor is used for creating this layer. This model can prove to be more accurate and light weight than other quantum counter parts as quantum convolutional neural network. Iterative approach is adopted for the minimization of the loss after each optimization the model is checked for accuracy and efficiency until the threshold is met. With advancement in quantum hardware the precision of such model can be increased.