Due to the investigation of the images in several parts of the life and the arising of the fast technology make the management of these images an open research area. Basically, the color feature considered as informative information that can be extracted from the image and help in improve the application performance. Based on the literature, this research found that there are several datasets that content images considered as a colorful images but some of these images content poor color information. For that, it’s unfair to treat all the dataset images as colorful images and this may lead to unsuccessful classification due to unfair color features that extracted from these images. To overcome this problem, this paper has proposed a color detector that can be used as a pre-processing stage to separate the dataset images into two classes colorful and colorless. The experiments have been carried out by using Caltech 101 dataset and the proposed method shows high level of discriminative power.