This paper proposes a parallel region growing image segmentation algorithm for Graphics Processing Units (GPU). It is inspired in a sequential algorithm widely used by the Geographic Object Based Image Analysis (GEOBIA) community. Initially, all image pixels are considered as seeds or primitive segments. Fine grained parallel threads assigned to individual pixels merge adjacent segments iteratively following a criterion, which seeks to minimize the average heterogeneity of image segments. Beyond spectral features the merging criterion considers morphological features, which can be efficiently computed in the underlying GPU architecture. Two algorithms using different merging criteria are proposed and tested. An experimental analysis upon five different test images has shown that the parallel algorithm may run more than 19 times faster than its sequential counterpart.