从Sylvain Ratabouil Android NDK(第2版)做一个例子,它从相机获取图像预览并原生处理,从YUV转换为RGB并应用滤色器。
代码非常简单,问题出现在传递给此函数的过滤器中:
public native void decode(Bitmap target, byte[] source, int filter);
目标是对ImageView的引用
来源是框架预览数据
过滤器是彩色滤镜
代码如下:
decode(mImageRed, data, 0xFFFF0000);
decode(mImageGreen, data, 0xFF00FF00);
decode(mImageBlue, data, 0xFF0000FF);
位图显示红色和蓝色交换,绿色没问题。
当我像这样交换红色和蓝色滤镜时:
decode(mImageRed, data, 0xFF0000FF);
decode(mImageGreen, data, 0xFF00FF00);
decode(mImageBlue, data, 0xFFFF0000);
*更改 0xFF0000FF 过滤器,其中 0xFFFF0000 用于红色图像,反之亦然。
在原生部分中,它所做的只是将过滤器与按位运算符和(&)一起应用:
bitmapContent[yIndex] &= pFilter;
有人知道颜色交换了吗?因为我认为0xFFFF0000是红色而不是0xFF0000FF。
这是解码功能:
void JNICALL
decode(JNIEnv *pEnv, jclass pClass, jobject pTarget, jbyteArray pSource, jint pFilter) {
// Retrieves bitmap information and locks it for drawing.
AndroidBitmapInfo bitmapInfo;
uint32_t *bitmapContent;
if (AndroidBitmap_getInfo(pEnv, pTarget, &bitmapInfo) < 0)
abort();
if (bitmapInfo.format != ANDROID_BITMAP_FORMAT_RGBA_8888)
abort();
if (AndroidBitmap_lockPixels(pEnv, pTarget, (void **) &bitmapContent) < 0)
abort();
// Accesses source array data.
jbyte *source = (*pEnv)->GetPrimitiveArrayCritical(pEnv, pSource, 0);
if (source == NULL)
abort();
int32_t frameSize = bitmapInfo.width * bitmapInfo.height;
int32_t yIndex, uvIndex, x, y;
int32_t colorY, colorU, colorV;
int32_t colorR, colorG, colorB;
int32_t y1192;
// Processes each pixel and converts YUV to RGB color.
// Algorithm originates from the Ketai open source project.
// See http://ketai.googlecode.com/.
for (y = 0, yIndex = 0; y < bitmapInfo.height; y++) {
colorU = 0;
colorV = 0;
// Y is divided by 2 because UVs are subsampled vertically.
// This means that two consecutives iterations refer to the
// same UV line (e.g when Y=0 and Y=1).
uvIndex = frameSize + (y >> 1) * bitmapInfo.width;
for (x = 0; x < bitmapInfo.width; x++, yIndex++) {
// Retrieves YUV components. UVs are subsampled
// horizontally too, hence %2 (1 UV for 2 Y).
colorY = max(toInt(source[yIndex]) - 16, 0);
if (!(x % 2)) {
colorV = toInt(source[uvIndex++]) - 128;
colorU = toInt(source[uvIndex++]) - 128;
}
// Computes R, G and B from Y, U and V.
y1192 = 1192 * colorY;
colorR = (y1192 + 1634 * colorV);
colorG = (y1192 - 833 * colorV - 400 * colorU);
colorB = (y1192 + 2066 * colorU);
colorR = clamp(colorR, 0, 262143);
colorG = clamp(colorG, 0, 262143);
colorB = clamp(colorB, 0, 262143);
// Combines R, G, B and A into the final pixel color.
bitmapContent[yIndex] = color(colorR, colorG, colorB);
bitmapContent[yIndex] &= pFilter;
}
}
(*pEnv)->ReleasePrimitiveArrayCritical(pEnv, pSource, source, 0);
if (AndroidBitmap_unlockPixels(pEnv, pTarget) < 0)
abort();
}
这里是如何分配位图的:
mImageR = Bitmap.createBitmap(size.width, size.height, Bitmap.Config.ARGB_8888);
mImageG = Bitmap.createBitmap(size.width, size.height, Bitmap.Config.ARGB_8888);
mImageB = Bitmap.createBitmap(size.width, size.height, Bitmap.Config.ARGB_8888);
mImageViewR.setImageBitmap(mImageR);
mImageViewG.setImageBitmap(mImageG);
mImageViewB.setImageBitmap(mImageB);
答案 0 :(得分:0)
您的decode()
功能是针对NV21视频格式进行硬编码的,但相机可以设置为YV12。这可以解释颜色'交换'。请注意,对于某些分辨率,YV12可以使用行填充(NV21保证“完全打包”)。
PS 在一次运行中生成三个平面可能要快得多,并且不清楚整数运算是否真的比现代CPU上的浮点快。
如果分辨率很高,处理时间很重要,请考虑使用renderscript。