The intricate tapestry of image mosaicing, while a cornerstone of computer vision, has long been constrained by the rigidity of classical, geometry-driven pipelines. Hand-crafted feature detectors and homography estimators, while effective in controlled environments, prove brittle when confronted with the challenges of real-world scenes: significant parallax, dynamic moving objects, repetitive textures, and illumination variance. This work introduces a fundamental paradigm shift, proposing a deep convolutional neural network (CNN) architecture designed for end-to-end, perceptually-driven image mosaic generation. Our model eschews the multi-stage, error-prone traditional workflow, instead learning a latent, correspondence-aware representation directly from image data. This network is trained not just to align pixels, but to synthesize a visually coherent scene, implicitly handling occlusions, warping artifacts, and exposure inconsistencies. Through extensive evaluation on benchmark and challenging real-world datasets, our CNN-based approach demonstrates a marked improvement in quantitative metrics and qualitative visual fidelity, particularly in high-parallax and dynamic scenarios. It successfully produces seamless, ghost-free mosaics where classical methods fail. This research establishes a new state-of-the-art, proving that a holistic, data-driven learning approach can master the complex art of scene assembly, paving the way for more robust and intelligent visual stitching applications in robotics, aerial imaging, and virtual reality.

🔒 This is a subscription article

Full text is available to subscribers and institutional members. Please choose an option below to access it.

1. Mulani AO, Patil RM, Liyakat KK. Discriminative appearance model for robust online multiple target tracking. Telematique. 2023;22(1):24–43.
2. Kumar MS, Ganesh D, Turukmane AV, Batta U, Sayyadliyakat KK. Deep convolution neural network based solution for detecting plant diseases. J Pharm Negat Results. 2022;13(1).
3. Liyakat KK. Significance and usage of face recognition system. Sch Res J Humanit Sci Engl Lang. 2017;4(20):4764–72.
4. Dixit AJ, Kazi MK. Iris recognition by Daugman’s method. Int J Latest Technol Eng Manag Appl Sci. 2015;4(6):90–3.
5. Liyakat KK. Significance of projection and rotation of image in color matching for high-quality panoramic images used for aquatic study. Int J Aquat Sci. 2018;9(2):130–45.
6. Kazi K. Reverse engineering’s neural network approach to human brain. J Commun Eng Syst. 2022;12(2):17–24.
7. Panwar V. A review on iris recognition system using machine and deep learning. In: Proc Int Conf Comput Commun Intell Syst (ICCCIS); 2022 Nov 4; India. IEEE; 2022. p. 857–66.
8. Shahdi SO, Abu-Bakar SA. Neural network-based approach for face recognition across varying pose. Int J Pattern Recognit Artif Intell. 2015;29(8):1556015.
9. Abd-Elhamied MR, Hashima WA, ElKateb S, Elhawary I, El-Geiheini A. Prediction of cotton yarn’s characteristics by image processing and ANN. Alexandria Eng J. 2022;61(4):3335–40.
10. Maniruzzaman M, Rahman MJ, Ahammed B, Abedin MM. Classification and prediction of diabetes disease using machine learning paradigm. Health Inf Sci Syst. 2020;8(1):7.
11. Mulani DA. Effect of rotation and projection on real time hand gesture recognition system for human computer interaction. J Gujrat Res Soc. 2019;21(16):3710–8.
12. Sreenivasulu MD, Devi JS, Arulprakash P, Venkataramana S, Kazi KS. Implementation of latest machine learning approaches for students grade prediction. Int J Early Child. 2022;14(3):3027–57.
13. Pereira F, Carvalho V, Soares F, Vasconcelos R, Machado J. Computer vision techniques for detecting yarn defects. In: Applications of Computer Vision in Fashion and Textiles. Woodhead Publishing; 2018. p. 123–45.
14. Hemamalini V, Rajarajeswari S, Nachiyappan S, Sambath M, Devi T, Singh BK, et al. Food quality inspection and grading using efficient image segmentation and machine learning-based system. J Food Qual. 2022;2022:5262294.
15. Meng Y, Wang R, Wang J, Yang J, Gui G. IRIS: Smart phone aided intelligent reimbursement system using deep learning. IEEE Access. 2019;7:165635–45.

The intricate tapestry of image mosaicing, while a cornerstone of computer vision, has long been constrained by the rigidity of classical, geometry-driven pipelines. Hand-crafted feature detectors and homography estimators, while effective in controlled environments, prove brittle when confronted with the challenges of real-world scenes: significant parallax, dynamic moving objects, repetitive textures, and illumination variance. This work introduces a fundamental paradigm shift, proposing a deep convolutional neural network (CNN) architecture designed for end-to-end, perceptually-driven image mosaic generation. Our model eschews the multi-stage, error-prone traditional workflow, instead learning a latent, correspondence-aware representation directly from image data. This network is trained not just to align pixels, but to synthesize a visually coherent scene, implicitly handling occlusions, warping artifacts, and exposure inconsistencies. Through extensive evaluation on benchmark and challenging real-world datasets, our CNN-based approach demonstrates a marked improvement in quantitative metrics and qualitative visual fidelity, particularly in high-parallax and dynamic scenarios. It successfully produces seamless, ghost-free mosaics where classical methods fail. This research establishes a new state-of-the-art, proving that a holistic, data-driven learning approach can master the complex art of scene assembly, paving the way for more robust and intelligent visual stitching applications in robotics, aerial imaging, and virtual reality.

🔒 This is a subscription article

Full text is available to subscribers and institutional members. Please choose an option below to access it.

1. Mulani AO, Patil RM, Liyakat KK. Discriminative appearance model for robust online multiple target tracking. Telematique. 2023;22(1):24–43.
2. Kumar MS, Ganesh D, Turukmane AV, Batta U, Sayyadliyakat KK. Deep convolution neural network based solution for detecting plant diseases. J Pharm Negat Results. 2022;13(1).
3. Liyakat KK. Significance and usage of face recognition system. Sch Res J Humanit Sci Engl Lang. 2017;4(20):4764–72.
4. Dixit AJ, Kazi MK. Iris recognition by Daugman’s method. Int J Latest Technol Eng Manag Appl Sci. 2015;4(6):90–3.
5. Liyakat KK. Significance of projection and rotation of image in color matching for high-quality panoramic images used for aquatic study. Int J Aquat Sci. 2018;9(2):130–45.
6. Kazi K. Reverse engineering’s neural network approach to human brain. J Commun Eng Syst. 2022;12(2):17–24.
7. Panwar V. A review on iris recognition system using machine and deep learning. In: Proc Int Conf Comput Commun Intell Syst (ICCCIS); 2022 Nov 4; India. IEEE; 2022. p. 857–66.
8. Shahdi SO, Abu-Bakar SA. Neural network-based approach for face recognition across varying pose. Int J Pattern Recognit Artif Intell. 2015;29(8):1556015.
9. Abd-Elhamied MR, Hashima WA, ElKateb S, Elhawary I, El-Geiheini A. Prediction of cotton yarn’s characteristics by image processing and ANN. Alexandria Eng J. 2022;61(4):3335–40.
10. Maniruzzaman M, Rahman MJ, Ahammed B, Abedin MM. Classification and prediction of diabetes disease using machine learning paradigm. Health Inf Sci Syst. 2020;8(1):7.
11. Mulani DA. Effect of rotation and projection on real time hand gesture recognition system for human computer interaction. J Gujrat Res Soc. 2019;21(16):3710–8.
12. Sreenivasulu MD, Devi JS, Arulprakash P, Venkataramana S, Kazi KS. Implementation of latest machine learning approaches for students grade prediction. Int J Early Child. 2022;14(3):3027–57.
13. Pereira F, Carvalho V, Soares F, Vasconcelos R, Machado J. Computer vision techniques for detecting yarn defects. In: Applications of Computer Vision in Fashion and Textiles. Woodhead Publishing; 2018. p. 123–45.
14. Hemamalini V, Rajarajeswari S, Nachiyappan S, Sambath M, Devi T, Singh BK, et al. Food quality inspection and grading using efficient image segmentation and machine learning-based system. J Food Qual. 2022;2022:5262294.
15. Meng Y, Wang R, Wang J, Yang J, Gui G. IRIS: Smart phone aided intelligent reimbursement system using deep learning. IEEE Access. 2019;7:165635–45.