Lately, orbital sensors of high spatial resolution are providing an increasing amount of data about the Earth surface. Analysis of these data implies in a high computational load, which has motivated researches on more efficient hardware and software for these applications. In this context, an important issue lies in the image segmentation that involves long processing times and is a key step in object based image analysis. The recent advances in modern programmable graphics units or GPUs have opened the possibility of exploiting the parallel processing capabilities to improve the segmentation performance. This work presents a parallel version of the multicriterion segmentation algorithm, introduced originally by Baatz and Schappe (2000), implemented in a GPU. The underlying hardware architecture consists of a massive parallel system with multiple processing elements designed especially for image processing. The parallel algorithm is based on processing each pixel as a different thread so as to take advantage of the fine-grain parallel capability of the GPU. In addition to the parallel algorithm, this dissertation also suggests a modification to the heterogeneity computation that improves the segmentation performance. The experiments under the proposed parallel algorithm present a speedup greater than 7 in relation to the sequential version.