A method of reducing ghost in images captured using a capsule endoscope while travelling in the gastrointestinal (GI) tract. The captured images contain ghost caused by reflections of multiple light sources by capsule housing of the capsule endoscope. The method derive, from the plurality of images, a ghost model comprising multiple ghost coefficients for relating light energies from the multiple light sources for a given image with ghost signals at multiple pixel locations for the given image. De-ghosted images are generated by compensating the plurality of images using estimated ghost signals based on derived ghost coefficients and the light energies from the multiple light sources. The process of deriving, from the plurality of images, the ghost model comprises removing any sensor gamma or any other non-linearity in pixel values of the plurality of images associated with the light energy.