You never mentioned that requirement. As long as you know the geo bounding box of the image, use the method in my previous reply to select the points within that rectangle and then

image = Import["~/Downloads/VIIRS-I5.png"];
GeoListPlot[selected, GeoBackground -> GeoStyling[{"Image", image}]]

I have run into the same issue when dealing with a large number of points. Filtering using GeoWithinQ is slow because it hits Wolfram's servers and will also fail for a large number of points. Filtering using GeoGridRange is also very slow. Outline of a workaround that I use.

geoPositions = RandomGeoPosition[Entity["Country", "UnitedStates"], 1000];

positions = geoPositions /. GeoPosition[p_] -> p; (* Extract coordinates *)

bounds = CoordinateBounds[positions]; (* Bounding box *)

(* Divide bounding box *)
lats = FindDivisions[First@bounds, 6];
longs = FindDivisions[Last@bounds, 6];

(* Subdivided bounding boxes *)    
grid = Table[{i, j}, {i, lats}, {j, longs}]; 

(* Rectangle for one subdivision *)
section = Rectangle[{35, -110}, {40, -100}];

(* Select positions in the rectangle *)    
selected = GeoPosition@Select[positions, RegionMember[section, #] &]

GeoListPlot[selected, 
 GeoBackground -> GeoStyling["SatelliteWithLabels"], GeoServer -> "DigitalGlobe"]