Matrix transformations are widely applied, particularly in image encryption, decryption, and network security, to weaken pixel correlation and enhance scrambling. Arnold’s 2D and 3D transforming matrices are commonly used to map image pixels into different coordinate systems with different bases. This paper introduces a multidimensional matrix incorporating rotation (r), power degree (α), and varying block sizes, tested on a database of images. To maintain a determinant of 1, matrix repair was applied. Lyapunov stability was analyzed for random variables, including block size, rotation (α), and power degree. Under high clockwise rotations, the matrix loses its ability to recover the original image through anticlockwise rotation, necessitating matrix repair to restore its characteristics and maintain the whirlpool-like transformation effect. Image simulations, along with metrics such as horizontal, vertical, and diagonal correlations and peak signal-to-noise ratio (PSNR), were used to evaluate the matrix’s effectiveness in reducing pixel correlation and ensuring after-transformation image retrieval.