溫馨提示×

您好，登錄后才能下訂單哦！

密碼登錄×
登錄注冊(cè)×
其他方式登錄
點(diǎn)擊 登錄注冊(cè) 即表示同意《億速云用戶服務(wù)條款》

C++ HLSL實(shí)現(xiàn)簡(jiǎn)單的圖像處理功能

發(fā)布時(shí)間：2020-09-24 06:30:35 來(lái)源：腳本之家 閱讀：269 作者：三米上 欄目：編程語(yǔ)言

由于對(duì)于dxva2解碼得到的數(shù)據(jù)不宜copy回內(nèi)存給CPU處理，所以最好的辦法是在GPU上直接進(jìn)行處理。D3D的像素著色器能夠?qū)ο袼刂苯舆M(jìn)行操作，實(shí)現(xiàn)點(diǎn)運(yùn)算極其簡(jiǎn)單方便，簡(jiǎn)單的卷積運(yùn)算效果也非常好。但D3D9的限制也很多，對(duì)于過(guò)于復(fù)雜的圖像處理則顯得有些不能勝任。

1.點(diǎn)運(yùn)算

點(diǎn)運(yùn)算用HLSL非常容易實(shí)現(xiàn)，幾乎是公式怎么寫，代碼就怎么寫。以RGB轉(zhuǎn)灰度圖顯示為例：

texture Tex0 ;
int iFlag = 0 ;
float aValue= 0.0 ;
float bValue= 0.0 ;
sampler2D YTex = 
sampler_state
{
  Texture = <Tex0> ;
  MipFilter = LINEAR ;
  MinFilter = LINEAR ;
  MagFilter = LINEAR ;

  AddressU = CLAMP ;
  AddressV = CLAMP ;
};
struct PS_INPUT
{
  float2 uvCoords0 : TEXCOORD0 ;
};
float4 Main( PS_INPUT input ) : COLOR0
{
  float4 yuvColor ;
  //rgb to gray 不知道是不是這么顯示的，姑且這么認(rèn)為
  float gray = tex2D( YTex, input.uvCoords0 ).r * 0.299 + tex2D( YTex, input.uvCoords0 ).g * 0.587 + tex2D( YTex, input.uvCoords0 ).b * 0.114 ;
  float s = 0 ;
  if(iFlag == 0)
  {
    s = aValue * gray + bValue/255 ;
  }
  else if(iFlag == 1)
  {
    s = aValue * log(1+gray) ;
  }
  else if(iFlag == 2)
  {
    s = aValue * pow(abs(gray),bValue) ;
  }
  yuvColor.r = s ;
  yuvColor.g = s ;
  yuvColor.b = s ;
  yuvColor.a = 1.0 ;
  return yuvColor ;
}

點(diǎn)運(yùn)算如此簡(jiǎn)單是因?yàn)镚PU是并行運(yùn)算的，我個(gè)人認(rèn)為可以看成是每一個(gè)像素點(diǎn)（BGRA）對(duì)應(yīng)一個(gè)線程，這大概就是OpenCL中所謂的數(shù)據(jù)并行。這是一個(gè)非常簡(jiǎn)單的程序，指令數(shù)少，程序結(jié)構(gòu)也很簡(jiǎn)單，shader 的版本用2.0就可以輕松編過(guò)。

2.卷積運(yùn)算舉例

指令數(shù)較多的情況2.0版本的shader就搞不定了，上3.0版本可以做一些簡(jiǎn)單的卷積運(yùn)算。以中值濾波為例：

texture Tex0 ;
matrix WorldMatrix;
matrix ViewMatrix;
matrix ProjMatrix;
sampler2D YTex = 
sampler_state
{
  Texture = <Tex0> ;
  MipFilter = LINEAR ;
  MinFilter = LINEAR ;
  MagFilter = LINEAR ;
  AddressU = CLAMP ;
  AddressV = CLAMP ;
};
struct VS_INPUT
 {
  float4 pos  : POSITION;
  float4 color : COLOR0;
  float2 tex  : TEXCOORD0;
 };
// 
struct VS_OUTPUT
 {
  float4 pos   : POSITION;
  float4 color  : COLOR0;
  float2 tex   : TEXCOORD0;
 };
float2 g_v4ScreenSize;
int ksize = 1 ;
float fLeft = -1.0f ;
float fTop = -1.0f ;
float fRight = -1.0f ;
float fBottom = -1.0f ;
//--------------------------------- BurTechnique --------------------------------------
VS_OUTPUT MainVS_Screen( VS_INPUT In )
{
  VS_OUTPUT Out = ( VS_OUTPUT )0;
  float4x4 matWorldView = mul(WorldMatrix,ViewMatrix);
  float4x4 matProject = mul(matWorldView,ProjMatrix);
  Out.pos = mul(In.pos,matProject);
  Out.tex = In.tex;
  Out.color = In.color;
  return Out;
}
float4 MainPS_Screen( VS_INPUT In ) : COLOR0
{
  float4 outColor = tex2D( YTex, In.tex ).rgba ;
  if(ksize <= 1 || ksize%2 == 0)
  {
    return outColor ;
  }
  if( ksize > 11 || ksize < 3)
  {
    return outColor ;
  }
  if(!(In.tex.x < fRight && In.tex.y < fBottom && In.tex.x > fLeft && In.tex.y > fTop))
  {
    return outColor ;
  }
  // 紋理大小
  float2 TexSize = float2( g_v4ScreenSize.x , g_v4ScreenSize.y );
  float x_off = 1.0f / TexSize.x;
  float y_off = 1.0f / TexSize.y;
  float2 fX0Y0 = In.tex - float2(x_off * ksize/2, y_off*ksize/2) ;
  float3 sum = {0.0f, 0.0f, 0.0f} ;
  if(ksize >= 3)
  {
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 0, y_off*0)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 0, y_off*1)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 0, y_off*2)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 1, y_off*0)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 1, y_off*1)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 1, y_off*2)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 2, y_off*0)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 2, y_off*1)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 2, y_off*2)).rgb;
  }
  if(ksize >= 5)
  {
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 3, y_off*0)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 3, y_off*1)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 3, y_off*2)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 3, y_off*3)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 3, y_off*4)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 4, y_off*0)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 4, y_off*1)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 4, y_off*2)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 4, y_off*3)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 4, y_off*4)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 0, y_off*3)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 1, y_off*3)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 2, y_off*3)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 0, y_off*4)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 1, y_off*4)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 2, y_off*4)).rgb;
  }
  if(ksize >= 7)
  {
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 5, y_off*0)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 5, y_off*1)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 5, y_off*2)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 5, y_off*3)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 5, y_off*4)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 5, y_off*5)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 5, y_off*6)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 6, y_off*0)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 6, y_off*1)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 6, y_off*2)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 6, y_off*3)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 6, y_off*4)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 6, y_off*5)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 6, y_off*6)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 0, y_off*5)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 1, y_off*5)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 2, y_off*5)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 3, y_off*5)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 4, y_off*5)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 0, y_off*6)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 1, y_off*6)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 2, y_off*6)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 3, y_off*6)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 4, y_off*6)).rgb;
  }
  if(ksize >= 9)
  {
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 7, y_off*0)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 7, y_off*1)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 7, y_off*2)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 7, y_off*3)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 7, y_off*4)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 7, y_off*5)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 7, y_off*6)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 7, y_off*7)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 7, y_off*8)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 8, y_off*0)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 8, y_off*1)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 8, y_off*2)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 8, y_off*3)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 8, y_off*4)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 8, y_off*5)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 8, y_off*6)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 8, y_off*7)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 8, y_off*8)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 0, y_off*7)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 1, y_off*7)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 2, y_off*7)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 3, y_off*7)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 4, y_off*7)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 5, y_off*7)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 6, y_off*7)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 0, y_off*8)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 1, y_off*8)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 2, y_off*8)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 3, y_off*8)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 4, y_off*8)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 5, y_off*8)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 6, y_off*8)).rgb;
  }
  if(ksize >= 11)
  {
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 9, y_off*0)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 9, y_off*1)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 9, y_off*2)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 9, y_off*3)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 9, y_off*4)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 9, y_off*5)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 9, y_off*6)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 9, y_off*7)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 9, y_off*8)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 9, y_off*9)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 9, y_off*10)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 10, y_off*0)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 10, y_off*1)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 10, y_off*2)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 10, y_off*3)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 10, y_off*4)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 10, y_off*5)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 10, y_off*6)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 10, y_off*7)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 10, y_off*8)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 10, y_off*9)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 10, y_off*10)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 0, y_off*9)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 1, y_off*9)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 2, y_off*9)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 3, y_off*9)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 4, y_off*9)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 5, y_off*9)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 6, y_off*9)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 7, y_off*9)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 8, y_off*9)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 0, y_off*10)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 1, y_off*10)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 2, y_off*10)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 3, y_off*10)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 4, y_off*10)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 5, y_off*10)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 6, y_off*10)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 7, y_off*10)).rgb;
    sum += tex2D( YTex , fX0Y0 + float2(x_off * 8, y_off*10)).rgb;
  }
  outColor = float4(sum/(ksize*ksize),1.0f);
  return outColor ;
}
//--------------------------- 技術(shù)---------------------------
technique BurTechnique
{
  pass P0
  {  
    LightEnable[0] = false;
    VertexShader = compile vs_3_0 MainVS_Screen();
    PixelShader = compile ps_3_0 MainPS_Screen();
  }
}

由于3.0版本的shader似乎不允許pixel shader單獨(dú)出現(xiàn)，所以我從點(diǎn)運(yùn)算用像素著色器實(shí)現(xiàn)改為用特效來(lái)實(shí)現(xiàn)。HLSL語(yǔ)法中有if語(yǔ)句，也有for語(yǔ)句，可是這個(gè)程序卻不厭其煩的把所有的都給列出來(lái)來(lái)，而沒有使用for循環(huán)。這是因?yàn)樵趯?shí)際使用中發(fā)現(xiàn)有一些限制，比如if語(yǔ)句的if（A>B）,A與B中必須有一個(gè)是常量，就像上面見到的那種形式；for循環(huán)中間的判斷也是如此，只是在第二層j循環(huán)中可以是第一層循環(huán)的i,即不可以

 for(int i=0;i<ksize;i++)
{
  for(int j=0;j<ksize1;j++)
    {
  ..........
    }
}

以上代碼的ksize與ksize1都必須為常數(shù)，例外的情況是ksize1可以為第一層循環(huán)的 i 。這個(gè)問(wèn)題不知道后續(xù)版本的shader有沒有，反正我當(dāng)前使用的版本有。

另外有一個(gè)需要注意的地方是指令數(shù)，2.0版本的shader支持的指令數(shù)相當(dāng)少，3.0版本則要多好多，我最長(zhǎng)寫到了400多條快500條時(shí)才導(dǎo)致編譯失敗。 還有一個(gè)需要提醒的是3.0版本的shader只支持D3D 9.0C以后的。如果要求做更為復(fù)雜的圖像處理，可以的話建議上D3D11，compute shader雖然我沒用過(guò)，但從介紹來(lái)說(shuō)，應(yīng)該可以處理一些更為復(fù)雜的圖像處理。

以上就是本文的全部?jī)?nèi)容，希望本文的內(nèi)容對(duì)大家的學(xué)習(xí)或者工作能帶來(lái)一定的幫助，同時(shí)也希望多多支持億速云！

向AI問(wèn)一下細(xì)節(jié)

免責(zé)聲明：本站發(fā)布的內(nèi)容（圖片、視頻和文字）以原創(chuàng)、轉(zhuǎn)載和分享為主，文章觀點(diǎn)不代表本網(wǎng)站立場(chǎng)，如果涉及侵權(quán)請(qǐng)聯(lián)系站長(zhǎng)郵箱：is@yisu.com進(jìn)行舉報(bào)，并提供相關(guān)證據(jù)，一經(jīng)查實(shí)，將立刻刪除涉嫌侵權(quán)內(nèi)容。

AI
助
手