In VIMz circuits, which implement various supported image transformations, array hashing serves as a fundamental primitive. During transformations applied row-by-row or region-by-region (e.g., in convolution-based editions), a running hash of both the original and transformed images is computed. This enables efficient operations, such as verification of proofs, directly on image commitments rather than the full images themselves.
However, hashing the entire image represents a significant portion of the circuit's size and computational overhead. This makes array hashing a considerable piece in terms of performance. Consequently, optimizing the hashing mechanism might be beneficial to improve overall circuit efficiency and scalability.