android照片美颜项目_Android OpenGLES 实时美颜(磨皮)的优化

在介绍实时美颜算法之前，你可以参考程序员杠把子的博客：

http://blog.csdn.net/oshunz/article/details/50536031

实时美颜算法考虑到性能的影响，PC平台上的很多美颜算法在手机镜头预览渲染的时候是很力不从心的。一般情况下都是基于高斯模糊进行实时美颜，这是性能和效果的折中方案。在程序员杠把子的博客中所描述的实时美颜算法实现的效果并不是很好，程序员杠把子的glsl是这样的:

precision mediump float;

varying mediump vec2 textureCoordinate;

uniform sampler2D inputImageTexture;

uniform vec2 singleStepOffset;

uniform mediump float params;

const highp vec3 W = vec3(0.299,0.587,0.114);

vec2 blurCoordinates[20];

float hardLight(float color)

{

if(color <= 0.5)

color = color * color * 2.0;

else

color = 1.0 - ((1.0 - color)*(1.0 - color) * 2.0);

return color;

}

void main(){

vec3 centralColor = texture2D(inputImageTexture, textureCoordinate).rgb;

blurCoordinates[0] = textureCoordinate.xy + singleStepOffset * vec2(0.0, -10.0);

blurCoordinates[1] = textureCoordinate.xy + singleStepOffset * vec2(0.0, 10.0);

blurCoordinates[2] = textureCoordinate.xy + singleStepOffset * vec2(-10.0, 0.0);

blurCoordinates[3] = textureCoordinate.xy + singleStepOffset * vec2(10.0, 0.0);

blurCoordinates[4] = textureCoordinate.xy + singleStepOffset * vec2(5.0, -8.0);

blurCoordinates[5] = textureCoordinate.xy + singleStepOffset * vec2(5.0, 8.0);

blurCoordinates[6] = textureCoordinate.xy + singleStepOffset * vec2(-5.0, 8.0);

blurCoordinates[7] = textureCoordinate.xy + singleStepOffset * vec2(-5.0, -8.0);

blurCoordinates[8] = textureCoordinate.xy + singleStepOffset * vec2(8.0, -5.0);

blurCoordinates[9] = textureCoordinate.xy + singleStepOffset * vec2(8.0, 5.0);

blurCoordinates[10] = textureCoordinate.xy + singleStepOffset * vec2(-8.0, 5.0);

blurCoordinates[11] = textureCoordinate.xy + singleStepOffset * vec2(-8.0, -5.0);

blurCoordinates[12] = textureCoordinate.xy + singleStepOffset * vec2(0.0, -6.0);

blurCoordinates[13] = textureCoordinate.xy + singleStepOffset * vec2(0.0, 6.0);

blurCoordinates[14] = textureCoordinate.xy + singleStepOffset * vec2(6.0, 0.0);

blurCoordinates[15] = textureCoordinate.xy + singleStepOffset * vec2(-6.0, 0.0);

blurCoordinates[16] = textureCoordinate.xy + singleStepOffset * vec2(-4.0, -4.0);

blurCoordinates[17] = textureCoordinate.xy + singleStepOffset * vec2(-4.0, 4.0);

blurCoordinates[18] = textureCoordinate.xy + singleStepOffset * vec2(4.0, -4.0);

blurCoordinates[19] = textureCoordinate.xy + singleStepOffset * vec2(4.0, 4.0);

float sampleColor = centralColor.g * 20.0;

sampleColor += texture2D(inputImageTexture, blurCoordinates[0]).g;

sampleColor += texture2D(inputImageTexture, blurCoordinates[1]).g;

sampleColor += texture2D(inputImageTexture, blurCoordinates[2]).g;

sampleColor += texture2D(inputImageTexture, blurCoordinates[3]).g;

sampleColor += texture2D(inputImageTexture, blurCoordinates[4]).g;

sampleColor += texture2D(inputImageTexture, blurCoordinates[5]).g;

sampleColor += texture2D(inputImageTexture, blurCoordinates[6]).g;

sampleColor += texture2D(inputImageTexture, blurCoordinates[7]).g;

sampleColor += texture2D(inputImageTexture, blurCoordinates[8]).g;

sampleColor += texture2D(inputImageTexture, blurCoordinates[9]).g;

sampleColor += texture2D(inputImageTexture, blurCoordinates[10]).g;

sampleColor += texture2D(inputImageTexture, blurCoordinates[11]).g;

sampleColor += texture2D(inputImageTexture, blurCoordinates[12]).g * 2.0;

sampleColor += texture2D(inputImageTexture, blurCoordinates[13]).g * 2.0;

sampleColor += texture2D(inputImageTexture, blurCoordinates[14]).g * 2.0;

sampleColor += texture2D(inputImageTexture, blurCoordinates[15]).g * 2.0;

sampleColor += texture2D(inputImageTexture, blurCoordinates[16]).g * 2.0;

sampleColor += texture2D(inputImageTexture, blurCoordinates[17]).g * 2.0;

sampleColor += texture2D(inputImageTexture, blurCoordinates[18]).g * 2.0;

sampleColor += texture2D(inputImageTexture, blurCoordinates[19]).g * 2.0;

sampleColor = sampleColor / 48.0;

float highPass = centralColor.g - sampleColor + 0.5;

for(int i = 0; i < 5;i++)

{

highPass = hardLight(highPass);

}

float luminance = dot(centralColor, W);

float alpha = pow(luminance, params);

vec3 smoothColor = centralColor + (centralColor-vec3(highPass))*alpha*0.1;

gl_FragColor = vec4(mix(smoothColor.rgb, max(smoothColor, centralColor), alpha), 1.0);

}

实时渲染的效果并不是非常理想，有些边缘地方在高度磨皮之后会出现比较明显的错误，像影子一样的东西，如下图所示：

MagicCamera的磨皮效果

从图中可以看到，高度磨皮的效果非常不自然，而且在光照比较强的情况下，边缘出现了非常明显的影子一样的透明的效果。

经过本人的研究，核心部分的高斯模糊并没有什么问题，但采样的数值并不对，采样的数值应该根据宽度和高度的变化进行相应的调整，经过调试优化，得到的GLSL代码如下：

precision lowp float;

uniform sampler2D inputTexture;

varying lowp vec2 textureCoordinate;

uniform int width;

uniform int height;

// 磨皮程度(由低到高: 0.5 ~ 0.99)

uniform float opacity;

void main() {

vec3 centralColor;

centralColor = texture2D(inputTexture, textureCoordinate).rgb;

if(opacity < 0.01) {

gl_FragColor = vec4(centralColor, 1.0);

} else {

float x_a = float(width);

float y_a = float(height);

float mul_x = 2.0 / x_a;

float mul_y = 2.0 / y_a;

vec2 blurCoordinates0 = textureCoordinate + vec2(0.0 * mul_x, -10.0 * mul_y);

vec2 blurCoordinates2 = textureCoordinate + vec2(8.0 * mul_x, -5.0 * mul_y);

vec2 blurCoordinates4 = textureCoordinate + vec2(8.0 * mul_x, 5.0 * mul_y);

vec2 blurCoordinates6 = textureCoordinate + vec2(0.0 * mul_x, 10.0 * mul_y);

vec2 blurCoordinates8 = textureCoordinate + vec2(-8.0 * mul_x, 5.0 * mul_y);

vec2 blurCoordinates10 = textureCoordinate + vec2(-8.0 * mul_x, -5.0 * mul_y);

mul_x = 1.8 / x_a;

mul_y = 1.8 / y_a;

vec2 blurCoordinates1 = textureCoordinate + vec2(5.0 * mul_x, -8.0 * mul_y);

vec2 blurCoordinates3 = textureCoordinate + vec2(10.0 * mul_x, 0.0 * mul_y);

vec2 blurCoordinates5 = textureCoordinate + vec2(5.0 * mul_x, 8.0 * mul_y);

vec2 blurCoordinates7 = textureCoordinate + vec2(-5.0 * mul_x, 8.0 * mul_y);

vec2 blurCoordinates9 = textureCoordinate + vec2(-10.0 * mul_x, 0.0 * mul_y);

vec2 blurCoordinates11 = textureCoordinate + vec2(-5.0 * mul_x, -8.0 * mul_y);

mul_x = 1.6 / x_a;

mul_y = 1.6 / y_a;

vec2 blurCoordinates12 = textureCoordinate + vec2(0.0 * mul_x,-6.0 * mul_y);

vec2 blurCoordinates14 = textureCoordinate + vec2(-6.0 * mul_x,0.0 * mul_y);

vec2 blurCoordinates16 = textureCoordinate + vec2(0.0 * mul_x,6.0 * mul_y);

vec2 blurCoordinates18 = textureCoordinate + vec2(6.0 * mul_x,0.0 * mul_y);

mul_x = 1.4 / x_a;

mul_y = 1.4 / y_a;

vec2 blurCoordinates13 = textureCoordinate + vec2(-4.0 * mul_x,-4.0 * mul_y);

vec2 blurCoordinates15 = textureCoordinate + vec2(-4.0 * mul_x,4.0 * mul_y);

vec2 blurCoordinates17 = textureCoordinate + vec2(4.0 * mul_x,4.0 * mul_y);

vec2 blurCoordinates19 = textureCoordinate + vec2(4.0 * mul_x,-4.0 * mul_y);

float central;

float gaussianWeightTotal;

float sum;

float sampler;

float distanceFromCentralColor;

float gaussianWeight;

float distanceNormalizationFactor = 3.6;

central = texture2D(inputTexture, textureCoordinate).g;

gaussianWeightTotal = 0.2;

sum = central * 0.2;

sampler = texture2D(inputTexture, blurCoordinates0).g;

distanceFromCentralColor = min(abs(central - sampler) * distanceNormalizationFactor, 1.0);

gaussianWeight = 0.09 * (1.0 - distanceFromCentralColor);