The conventional Gabor Transforms, Complete and Generalized cannot be appliedfor image matrices received as different partitions. In practice, very large images are received in the form of submatrix partitions. Then, taking transform on different partitions (of the same image) at different times is of prime importance. The situation can be easily handled i f the transformation algorithm is made recursive. In this paper we present a Recursive Gabor Transform (RGT) amenable for pipelining panitions of diflerent images interleaved. Apriori knowledge of image size is not required for the RGT unlike the conventional Gabor Transform. A dedicated rnassively parallel architecture based on Pacube VLSI arrays is presented which achieves a high degree of pipelining with respect to different image matrices. The pipelining rate is proportional to the delay of a Pacube macrocell in mode I operation.where T, T > 0. When 'T = T, the critical sampling case (1) is the Comp1el.e Gabor Transform (CGT). When T c T', the oversampling case, (1) is the GGT, g is the window function. The matrix formulation proposed by Yao [ 11, for GGT can be expressed as f = G a ... (2) where f = (fo, f,, ..., fKM.I)T and a = (a, , , a,, ..,, +M.I)T We can then decompose the coefficient vector a = (IK * E,) D 1 f ... (3) where I, is the Kronecker product, the form is a matrix of euv = exp (-2mv/M'), U, v = 0,1, ,.. , M'-1