Taking a look at your particular image, I think you might have some success if you use ColorTogray to split the image up into it’s red, green and blue components (RBG). Some colors are enhanced in a particular channel and you can use just that channel for identifying the features you are interested in.
For example, the brown DAB seems to be prominent in the blue channel, so you can run IdentifyPrimAutomatic on the blue channel to get the DAB (you should invert it first using InvertIntensity, since IdentifyPrim assumes you are looking for light objects on a dark background).
Similarly, the green channel seems to enhance both the brown and purple nuclei. So you can use InvertIntensity on the green channel (to get the bright objects), then use IdentifyPrimAutomatic on the result to get all the nuclei.
Since you want just the purple ones, you can (a) use MeasureObjectIntensity on the nuclei above using the inverted blue channel as your image, then (b) use FilterByObjectMeasurements using the mean intensity (Intensity category, feature number 2) with an appropriate threshold.
The reason this works is that you are taking advantage of the fact that you have identified all the nuclei but in the inverted blue image, only the brown nuclei are bright. FilterByObjectMeasurements allows you set a threshold of maximum brightness that will exclude the brown nuclei, leaving the purple ones.
Hope this helps,