Seeded region growing has two main parameters, namely seed initialization and threshold determination. Parameter values can be done manually or automatically. Automatic parameter assessment is suitable to be applied on a real time basis, but the parameter assessment must be precise. This is due to the greater risk of segmentation errors than manual. This research aims to propose automatic determination of seeded region growing in the watershed region for image segmentation of tuna. The tuna image was processed into the hue, saturation, and intensity (HSI) color space. The hue color space was then taken up in the formation of the watershed region. After that, the density of a region was calculated, then the density was sorted and the highest density was taken. The region that had the highest density was taken based on the highest gray level intensity, then the threshold was obtained from the difference between the average number of regions used and the average intensity of the region left. The results of tuna image segmentation were successfully carried out by proving the average values of relative foreground area error (RAE), missclassification error (ME), and modified Hausdroff distance (MHD) respectively for category 1 data of 6.77%, 1.78% and 0.18%, while for category 2 data of 3.44%, 1.30% and 0.66%.