Implementation d'un processeur video sur FPGA pour la detection et la correction de reflexions speculaires dans des images endoscopiques

The framework of the project is surgery. When operating a patient, the surgeon can either proceed by an invasive surgery or a minimally invasive surgery. The latter is made by endoscopy. It consists in inserting surgical instruments through cuts inside the patient's body. A small camera, the endoscope, is also inserted to guide the surgeon throughout the operation.;The endoscopic surgery has many advantages. In fact, it causes less pain, reduces the length of hospitalization, and allows for a quick recovery. However, it has some disadvantages for the surgeon. The endoscopic surgery enables solely a two-dimensional vision. Though, a three dimensions vision is more appropriate. To alleviate this constraint, research has been done to create an augmented reality surgical environment, an environment in which a virtual 3D model of the organ is superimposed on the real model. Therefore, this overlay requires an accurate segmentation of the objects. Note that, the light source of the endoscope, being too close to the organs and tools, reflects some luminous artefacts that damage useful information and prevent an accurate segmentation.;The aim of the project is to find out the specular reflections in a video sequence and to correct them. These two steps must occur in real time in order to transmit the video footage in a clear and a transparent way to the surgeon. The FPGA, a hardware architecture, is used to achieve these real time objectives.;Two detection methods are used to locate the position of the specular reflections. The first method consists in using a color histogram to detect the minimal intensity pixel of these reflections. The second method uses simultaneously two plans (saturation and intensity) of a frame. This method consists in making a threshold on both the saturation and intensity plans in order to isolate the specular area. The principle of image inpainting is used as the correction method. Indeed, it collects information around the area to correct, and spreads it inside that same area. To cope of with limited resources of the FPGA, several enhancements are proposed to implement these methods. The real-time objectives are achieved. The system detects and corrects specularities within 0.8 ms and uses 91% of the available LUTs.