We present a new optical image compression approach based on spectral selection and using a segmented filter. In order to identify the method offering the best quality of reconstruction of the optically compressed images, we investigated several spectral segmentation methods. The compression is carried out in the spectral domain and results in a compound spectrum referred to as the segmented spectrum. Indeed, relevant information should be selected from each spectrum to form a unique spectrum that is able to reconstruct the respective images. To separate the images in the output plane, after applying a second Fourier transform, the segmented spectrum should include a distinct carrier for each spectrum. Information selection consists of comparing pixel by pixel the spectra of the images being compressed. To identify the winner among the spectral pixels, we investigated four forms of information, namely the intensity of the pixel in question, its complex gradient, its phase gradient and its real part. In addition to these forms, we considered three options of selection, namely selection of the pixel with the most important amount of information, selection with respect to a fixed probability function and selection based on a matched probability function.