OCL code fixed, fix for NEON added

modules/imgproc/src/color.cpp

@@ -6608,6 +6608,14 @@ struct RGB2Luv_f
             {
                 float32x4x3_t v_src = vld3q_f32(src);
 
+                v_src.val[0] = vmaxq_f32(v_src.val[0], vdupq_n_f32(0));
+                v_src.val[1] = vmaxq_f32(v_src.val[1], vdupq_n_f32(0));
+                v_src.val[2] = vmaxq_f32(v_src.val[2], vdupq_n_f32(0));
+
+                v_src.val[0] = vminq_f32(v_src.val[0], vdupq_n_f32(1));
+                v_src.val[1] = vminq_f32(v_src.val[1], vdupq_n_f32(1));
+                v_src.val[2] = vminq_f32(v_src.val[2], vdupq_n_f32(1));
+
                 if( gammaTab )
                 {
                     v_src.val[0] = vmulq_f32(v_src.val[0], vdupq_n_f32(gscale));
@@ -8574,7 +8582,7 @@ static bool ocl_cvtColor( InputArray _src, OutputArray _dst, int code, int dcn )
                                coeffs[j] + coeffs[j + 1] + coeffs[j + 2] < 1.5f*(lab ? LabCbrtTabScale : 1) );
                 }
 
-                float d = 1.f/(_whitept[0] + _whitept[1]*15 + _whitept[2]*3);
+                float d = 1.f/std::max(_whitept[0] + _whitept[1]*15 + _whitept[2]*3, FLT_EPSILON);
                 un = 13*4*_whitept[0]*d;
                 vn = 13*9*_whitept[1]*d;
 
@@ -8641,9 +8649,9 @@ static bool ocl_cvtColor( InputArray _src, OutputArray _dst, int code, int dcn )
                 coeffs[i+bidx*3] = _coeffs[i+6] * (lab ? _whitept[i] : 1);
             }
 
-            float d = 1.f/(_whitept[0] + _whitept[1]*15 + _whitept[2]*3);
-            un = 4*_whitept[0]*d;
-            vn = 9*_whitept[1]*d;
+            float d = 1.f/std::max(_whitept[0] + _whitept[1]*15 + _whitept[2]*3, FLT_EPSILON);
+            un = 4*13*_whitept[0]*d;
+            vn = 9*13*_whitept[1]*d;
 
             Mat(1, 9, CV_32FC1, coeffs).copyTo(ucoeffs);
         }

modules/imgproc/src/opencl/cvtcolor.cl

@@ -1963,6 +1963,10 @@ __kernel void BGR2Luv(__global const uchar * srcptr, int src_step, int src_offse
 
                 float R = src[0], G = src[1], B = src[2];
 
+                R = clamp(R, 0.f, 1.f);
+                G = clamp(G, 0.f, 1.f);
+                B = clamp(B, 0.f, 1.f);
+
 #ifdef SRGB
                 R = splineInterpolate(R*GammaTabScale, gammaTab, GAMMA_TAB_SIZE);
                 G = splineInterpolate(G*GammaTabScale, gammaTab, GAMMA_TAB_SIZE);
@@ -2067,15 +2071,21 @@ __kernel void Luv2BGR(__global const uchar * srcptr, int src_step, int src_offse
                 __global const float * src = (__global const float *)(srcptr + src_index);
                 __global float * dst = (__global float *)(dstptr + dst_index);
 
-                float L = src[0], u = src[1], v = src[2], d, X, Y, Z;
-                Y = (L + 16.f) * (1.f/116.f);
-                Y = Y*Y*Y;
-                d = (1.f/13.f)/L;
-                u = fma(u, d, _un);
-                v = fma(v, d, _vn);
-                float iv = 1.f/v;
-                X = 2.25f * u * Y * iv;
-                Z = (12 - fma(3.0f, u, 20.0f * v)) * Y * 0.25f * iv;
+                float L = src[0], u = src[1], v = src[2], X, Y, Z;
+                if(L >= 8)
+                {
+                    Y = fma(L, 1.f/116.f, 16.f/116.f);
+                    Y = Y*Y*Y;
+                }
+                else
+                {
+                    Y = L * (1.0f/903.3f); // L*(3./29.)^3
+                }
+                float up = 3.f*fma(L, _un, u);
+                float vp = 0.25f/fma(L, _vn, v);
+                vp = clamp(vp, -0.25f, 0.25f);
+                X = 3.f*Y*up*vp;
+                Z = Y*fma(fma(12.f*13.f, L, -up), vp, -5.f);
 
                 float R = fma(X, coeffs[0], fma(Y, coeffs[1], Z * coeffs[2]));
                 float G = fma(X, coeffs[3], fma(Y, coeffs[4], Z * coeffs[5]));
@@ -2129,14 +2139,20 @@ __kernel void Luv2BGR(__global const uchar * src, int src_step, int src_offset,
                 float L = src[0]*(100.f/255.f);
                 float u = fma(convert_float(src[1]), 1.388235294117647f, -134.f);
                 float v = fma(convert_float(src[2]), 1.027450980392157f, - 140.f);
-                Y = (L + 16.f) * (1.f/116.f);
-                Y = Y*Y*Y;
-                d = (1.f/13.f)/L;
-                u = fma(u, d, _un);
-                v = fma(v, d, _vn);
-                float iv = 1.f/v;
-                X = 2.25f * u * Y * iv ;
-                Z = (12 - fma(3.0f, u, 20.0f * v)) * Y * 0.25f * iv;
+                if(L >= 8)
+                {
+                    Y = fma(L, 1.f/116.f, 16.f/116.f);
+                    Y = Y*Y*Y;
+                }
+                else
+                {
+                    Y = L * (1.0f/903.3f); // L*(3./29.)^3
+                }
+                float up = 3.f*fma(L, _un, u);
+                float vp = 0.25f/fma(L, _vn, v);
+                vp = clamp(vp, -0.25f, 0.25f);
+                X = 3.f*Y*up*vp;
+                Z = Y*fma(fma(12.f*13.f, L, -up), vp, -5.f);
 
                 float R = fma(X, coeffs[0], fma(Y, coeffs[1], Z * coeffs[2]));
                 float G = fma(X, coeffs[3], fma(Y, coeffs[4], Z * coeffs[5]));