HeliNet is a project recently funded by the European Community, within the Fifth Framework Programme of research and technological development, for the design of a network of aerodynamic stratospheric platforms. This network is intended to yield integrated services of broadband communications, as well as optical surveillance and traffic monitoring. Particularly for the surveillance application, an optical payload will be mounted on the platforms, so enabling for the analysis and processing of high resolution digital pictures of the ground. In this paper we deal with image processing algorithms to be run on the images acquired by the on-board optical device, addressing specific features of the HeliNet platforms. In particular, HeliNet application development is constrained by platform requirements on the available payload power and weight, which raise the need for low complexity algorithms; moreover, platform motion and low flight height put the issue of available image resolution in a light potentially more promising with respect to other spaceborne systems. We assess general requirements for image processing applications, including motion and vibration compensation by efficient image registration, image restoration (motion blur compensation) and resolution enhancement (super-resolution). As the limited bandwidth of the transmissive interface strongly constrains the amount of data which can be transmitted to ground, it is desirable that some image processing is performed on board with limited computational complexity. Therefore we describe the requirements of intelligent data compression and processing algorithms to be run on board. These algorithms will be designed according to the HeliNet specific constraints and foreseen applications, having low complexity, and permitting to select and interpret images/data of interest, which are the only ones transmitted to ground, thus fulfilling the available data rate of the communication interface. We show that this approach is successful in trading-off between limited bandwidth and quality of processing results.