Resampling is done when you reproject a dataset to a new grid (such as a new projection system or when resizing an image). Both bilinear and cubic convolution resampling perform a type of interpolation of the pixels in a neighborhood to provide the new output pixel value. This type of resampling is useful for providing a smoothed looking image after it is reprojected but it does change the original radiance/reflectance values. Nearest neighbor resampling does not perform any interpolation and retains the original values in the new grid. When you display a file in ENVI, there is typically a default stretch that is used to improve image contrast. The default stretch is a 2% linear stretch. This stretch is only for display purposes and does not change the data values of the image. When you bring up the cursor location/value tool, you can see two sets of values. The 'Data' values are the actual image pixel values. The 'Scrn' values are the stretched values that are used to display the image on your monitor and can only range from 0-255 for RGB. Applying a stretch using the Enhance menu from a display only changes the Scrn values and not the Data values.