In this paper, we propose a quantum-classical hybrid framework for binary classification of
handwritten digits. We encode images as quantum states and apply sequential unsharp measurements to extract features while preserving partial coherence. These features are used as input of a classical neural network which is then trained to distinguish between digits, specifically, ‘0’ and ‘4’ from the MNIST dataset. This model demonstrate 94.60 % binary classification accuracy further proving that the unsharp measurement is a valid feature extraction technique.
handwritten digits. We encode images as quantum states and apply sequential unsharp measurements to extract features while preserving partial coherence. These features are used as input of a classical neural network which is then trained to distinguish between digits, specifically, ‘0’ and ‘4’ from the MNIST dataset. This model demonstrate 94.60 % binary classification accuracy further proving that the unsharp measurement is a valid feature extraction technique.