Abstract: In this study, we propose a novel segmentation necessity metrics with the considerations of the implicit region sizes or image complexity. As for the region-based image processing such as region-based image retrieval, classification, pattern recognition or computer vision, one image should be segmented into different size and shape regions according to the colors or edge information of the image. However, for some images with great complexity or without meaningful implicit regions, these images should avoid segmentation, as the images with larger numbers of fragmentary regions would consume a great deal of computation or storages. It is necessary to exclude these images before segmentation. According the segmentation necessity metric, the minimum region sizes of the incoming image analysis, retrieval or object recognition is used to evaluate the values of segmentation necessity metrics. This is also the unique peculiarity of the metric when compared with other segmentation necessity metrics, e.g., wavelet modulus maxima points-based or connectivity index-based approaches. An image database with many categories is constructed to test the proposed segmentation necessity metric and other common approaches. Experiment results show the proposed metric could keep consistent with the implicit region sizes and numbers, while the numbers have directed relations with image segmentation complexity. It`s also verified that the proposed scheme can achieve more verdict accuracy than other schemes and could achieve the verdict accuracy of 98.5% at the requirement region sizes of 50 pixels.