如何优化此haskell功能

时间:2014-01-13 15:40:38

标签: haskell optimization

我需要在调色板ps中找到与给定颜色p最接近的颜色。如何在不更改类型的情况下尽快使函数nearestColor Pixel8PixelRGB8。到目前为止,我已尝试内联。

import qualified Data.Vector as V

type Pixel8 = Word8    

data PixelRGB8 = PixelRGB8 {-# UNPACK #-} !Pixel8 -- Red
                           {-# UNPACK #-} !Pixel8 -- Green
                           {-# UNPACK #-} !Pixel8 -- Blue
           deriving (Eq, Ord, Show)

nearestColor :: PixelRGB8 -> Vector PixelRGB8 -> PixelRGB8
nearestColor p ps = snd $ V.minimumBy comp ds
  where
    ds = V.map (\px -> (dist2Px px p, px)) ps
    comp a b = fst a `compare` fst b

dist2Px :: PixelRGB8 -> PixelRGB8 -> Int
dist2Px (PixelRGB8 r1 g1 b1) (PixelRGB8 r2 g2 b2) = dr*dr + dg*dg + db*db
  where
    (dr, dg, db) =
      ( fromIntegral r1 - fromIntegral r2
      , fromIntegral g1 - fromIntegral g2
      , fromIntegral b1 - fromIntegral b2 )

1 个答案:

答案 0 :(得分:6)

如果您想使用单个调色板并请求不同颜色,我首先要翻转您的签名:

type Palette = V.Vector PixelRGB8
nearestColor :: Palette -> PixelRGB8 -> PixelRGB8

这有助于部分应用,并允许对调色板配置进行记忆。

接下来,您要这样做:将调色板重新存储在适合快速查找的数据结构中。由于你基本上对ℝ 3 中的欧几里德距离感兴趣(BTW不是真正理想的颜色比较),这是一个非常普遍的问题。经典结构是 k -d树,has long been used for such a nearest-neighbour search。确实有a Haskell library可用,这对您来说非常方便:

import qualified Data.Trees.KdTree a s KD

instance KD.Point PixelRGB where
  dimension _ = 3
  coord 0 (PixelRGB r _ _) = fromIntegral r
  coord 1 (PixelRGB _ g _) = fromIntegral g
  coord 2 (PixelRGB _ _ b) = fromIntegral b
  dist2 = fromIntegral . dist2Px

然后我们可以将调色板转换为这样的树:

type FastPalette = KD.KdTree PixelRGB8
accelPalette :: Palette -> FastPalette
accelPalette = KD.fromList . V.toList

最后只使用库提供的下一个邻居搜索:

nearestColor palette = fromJust . KD.nearestNeighbor fpal
 where fpal = accelPalette palette