IMAGE CONTRAST IMPROVEMENT: COMPARATIVE ANALYSIS OF TRADITIONAL SOFTWARE TOOLS AND MIRNET NEURAL NETWORK

Abstract

The purpose of this article is to compare the effectiveness of traditional Adobe Photoshop and Adobe Photoshop Lightroom tools with MIRNet neural network in terms of contrast enhancement tasks, evaluating the results by quantitative metrics and visual quality on real datasets. It is noted that professional photographers continue to rely on graphic editors and programs (in particular, Adobe Photoshop and Adobe Photoshop Lightroom) for manual image processing. The development of deep learning has led to the emergence of neural networks and architectures that are directly trained on pairs of low-quality/high-quality images and are capable of performing nonlinear mapping of pixel values and general detail restoration. Among them, MIRNet neural network occupies a special place. A brief review of the main sources indicates that in recent years, there has been a significant increase in the number of works devoted to image enhancement using deep neural network models. However, there are virtually no direct comparisons between Adobe Photoshop and Adobe Photoshop Lightroom and modern neural networks. The results show that all Adobe Photoshop and Adobe Photoshop Lightroom tools improve image quality, but their effectiveness varies significantly. The study found that traditional graphic editor and software tools (Adobe Photoshop, Adobe Photoshop Lightroom) remain reliable and flexible for local and global corrections, but their effectiveness depends on manual user interaction and does not always guarantee optimal structural compliance. It has been determined that the MIRNet neural network is capable of performing adaptive multiscale correction without aggressively stretching the tonal range, thereby preserving the naturalness of the scene and minimizing the risk of distortion. A comparison using the objective SSIM metric confirmed this trend: MIRNet demonstrates higher SSIM values (>0.99) during contrast correction. Thus, the MIRNet neural network has the potential to become the standard for image enhancement in tasks where automation, stability, and reliability are required.