Automatic Image clipping of Product Photos

Hey there, We are looking for someone who can solve our Task for automatic Image clipping. For the Project we prefer Java as well as OpenCV. (still we are open for other/better Ideads) We want to clip our Images automaticly. We take the Photos on a grey Flor (around 120,120,120) in two Perspektives. One from the Top the othere from the Side. The goal is to place all the taken Photos in one directory and let the Software run over night, so we have the clipped photos ready to use the next day. We dont need any kind of GUI just a runnable software. I already experimented with grabcut and so on. My results were acutually pretty good, but still far from the Quality we want. This programm only needs to work for carpets. Normal rectengural-size, round and runners. If the carpet is not 100% in Line we'd like to have that also fixed for runner and rectengural-size with .transform or something. In my opinion, this Task shouldnt be to difficult. Still there is the tricky part of the carpet fringe. Some carpets have them, some dont. Some are even in an geryish color. We take our picutes in 6000x4000. The resulting carpets we want in two Sizes. First original Size and second with 1500px in height and variable width(depending on the width of the product). What Type of Image Input do you prefer or recommend? At the moment we take picutres in RAW .NEF. As Outputtpe we would like to have JPG. Size of the JPG should be as small as possible and we dont want to lose any quality. Do you have any Suggestions? Since I program with Java myself, i would like to get the Source-Code of this Projekt and not a Version that is only runnable. What do you think how much time will it you to complete this task? Last but not least the attached images show before and after. (in this case two diffrent products) Of course the Software should work for many diffrent kind of carpets. If you are interested, we will send you the needed Pictures to test. The software should work for 99 out of 100 carpets, so 99% of the time. Best regards