A colored three-dimensional point cloud is encoded in at least two depth and texture images. The proposed method and device use inverse project pixels of the first image and inverse project pixels to the second image. When the inverse projecting of a first pixel of the first image and the inverse projecting of a second pixel of the second image refer to a common generated point, that is two 3D points closer to each other than the imprecision due to the encoding and compression of the images, the color value of the pixel of first and second pixels having the lowest local depth variance is attributed to the common generated point.