垃圾收集不适用于Golang中的[] gocv.Mat

Assume that the img is a 3 channel Mat in the code. There is a memory leak with the code. I guess that the pointers(reference) in the slice "matsplits" are not deleted with garbage-collection in go. How can I fix it?

for{
    matsplits := gocv.Split(img)
    matsplits[0].Close()
    matsplits[1].Close()
    matsplits[2].Close()
}

Kind of above codes cause memory leak. I am sure that the Mat objects in imgarr are closed, but the memory usage is still growing up. Why?

Update: part of codes in my project

processed := 0
for processed < proc.imgNumber {
    grayhconcatImg := <-proc.processedImg[0][chindex]
    var roiList roilist
    var numStartPosList numStartPos
    for x := 0; x < 11520-w; x++ {
        test := gocv.NewMat()
        testRegion := grayhconcatImg.img.Region(image.Rect(x, 0, x+w, h))
        gocv.BitwiseXor(chimg, testRegion, &test)
        testRegion.Close()
        //testsplit := gocv.Split(test)
        test.Close()
        //testsplit[0].Close()
        //testsplit[1].Close()

    processed++
}

The memory leak occurs if "testsplit"'s are unmarked. len(testsplit) is 2. I have checked that testsplit[0] and testsplit[1] have been closed correctly after testsplit[i].Close().

After Close(), the memory will be clear partly after the gc comes. Check it like

    matsplits := gocv.Split(img)
    matsplits[0].Close()
    matsplits[1].Close()
    matsplits[2].Close()
    runtime.GC()

exec like GODEBUG=gctrace=1 go run main.go 2>xx.log and you can find the what gc actual do in xx.log