Intra-User Image Ranking Using Computation Efficient Quantum Convolutional Neural Networks for Selecting Highly Relevant and Non-Redundant Images
DOI:
https://doi.org/10.70917/ijcisim-2026-2731Keywords:
Computation-efficient Quantum Convolutional Neural Networks, Dynamic Binary Swordfish Movement Optimization Algorithm, Deep feature embedding, Redundancy reduction, Pattern recognition, Visual similarity learningAbstract
Intra-user image ranking refers to the process of organizing and prioritizing images within a single user’s collection
based on their relevance and quality.The rapid growth of user-generated images has created a major challenge of organizing
and identifying the most meaningful images within a single user’s collection as low-quality and lack popularity information
such as likes and views. To overcome these issues,Intra-User Image Ranking Using Computation-Efficient Quantum
Convolutional Neural Networks for Selecting Highly Relevant and Non-Redundant Images (IUIR-CEQCNN-HRNI) is
proposed.Computation-Efficient Quantum Convolutional Neural Networks (CEQCNN) captures complex image patterns
using fewer parameters by handling high-dimensional image effectively.The method is used to select most relevant and non
relevant images from personal image collections.These features are then encoded into quantum states using angle encoding
allowing multiple data values through quantum superposition. The QCNN applies parameterized quantum convolution layers
composed of small entangled qubit circuits like convolution layers to capture local patterns, followed by quantum pooling
operations that reduce number of qubits and retain most informative quantum states.Dynamic Binary Swordfish Movement
Optimization Algorithm (DBSMOA) is used to solve optimization problems based on swarms where the hunting process of
swordfish is simulated. It uses adaptive techniques of exploration and exploitation to dynamically evolve the candidate
solutions and maps the movement to a binary decision process to the best possible subset. The model is tested using the Flickr
image data set and the results show the model with 99.41% accuracy, 98.41% recall and 98.72% f1-score outperforming
existing techniques by a significant margin. The proposed IUIR-CEQCNN-HRNI method provides reduced redundancy,
improved image quality selection, higher ranking accuracy and efficient management of large personal image collections.
