inline void operator+=(const histogram& hist)
{
const_iterator it = hist.begin(), e = hist.end();
for (; it != e; ++it)
update(it->first, it->second);
}
//! apply k-means classify to histogram
void apply_k_means_classify(const histogram<double>& input,
const std::vector<double>& mu,
std::vector<int>& cls )
{
int number_of_classes=mu.size();
int i,j,k;
//calculate all the classes
for(j=0;j<input.size();j++)
{
int best_k=0;
double best_dist=0;
for(k=0;k<number_of_classes;k++)
{
double dist=fabs(input.value(j)-mu[k]);
if(dist<best_dist|| best_k==0)
{
best_dist=dist;
best_k=k+1;
}
}
cls[j]=best_k;
}
}
double ks_distance(const histogram<double>& sample1,const histogram<double>& sample2)
{
//let's include the whole range
double _min=std::min(sample1.min(),sample2.min());
double _max=std::max(sample1.max(),sample2.max());
double _range=_max-_min;
histogram<double> _s1(sample1);
histogram<double> _s2(sample2);
_s1.convert_to_commulative();
_s2.convert_to_commulative();
int _size=std::max(_s1.size(),_s2.size())*2;
double _bin=_range/_size;
double dist=0.0;
double v=0.0;
for(int i=0;i<_size;i++,v+=_bin)
{
double d=fabs(_s1[v]-_s2[v]);
if(d>dist) dist=d;
}
return dist;
}
double kl_distance(const histogram<double>& sample1,const histogram<double>& sample2)
{
//let's include the whole range
double _min=std::min(sample1.min(),sample2.min());
double _max=std::max(sample1.max(),sample2.max());
double _range=_max-_min;
//now we are going to iterate through samples
double distance=0.0;
int _size=std::max(sample1.size(),sample2.size());
double _bin=_range/_size;
double v=_min+_bin/2.0;
for(int i=0;i<_size;i++,v+=_bin)
{
double _P=0.0;
if(v>=sample1.min() && v<=sample1.max()) _P=sample1[v];
double _Q=0.0;
if(v>=sample2.min() && v<=sample2.max()) _Q=sample2[v];
if(_P>0.0 && _Q>0.0)//TODO: epsilon?
distance+=_P*log(_P/_Q);
}
return distance;
}
/* Add a other histogram to this one.
* Does nothing if histogram parameters ( min, max, num_buckets ) don't match.
*/
void accumulate( const histogram& other )
{
if ( min() != other.min() || max() != other.max() || data().size() != other.data().size() )
return;
for ( size_t j = 0, num_buckets = _data.size(); j < num_buckets; ++j )
_data[ j ] += other.data()[ j ];
_num_entries += other.num_entries();
}
void compute_histogram(unsigned short *depthimage,
int npts,
histogram &hist,
bool normalize)
{
hist.clear();
for (int i = 0; i < npts; i++)
hist.insert_point(depthimage[i]);
if (normalize)
hist.normalize();
}
void checkForUniq(const vector<string>& matrix, histogram& hist, vector<Block>& uniqs,
int rowSize, int colSize, int numRows, int numCols) {
int x = numRows - rowSize + 1;
int y = numCols - colSize + 1;
for(int i=0;i<x;++i) {
for(int j=0;j<y;++j) {
hist.init();
hist.evaluate(matrix, i, j, rowSize, colSize);
if(hist.isUnique()) {
Block blk(i, j, rowSize, colSize);
uniqs.push_back(blk);
}
}
}
}
static
python::dict
histogram_array_interface(histogram& self) {
python::dict d;
python::list shape;
for (unsigned i = 0; i < self.dim(); ++i)
shape.append(self.shape(i));
if (self.depth() == sizeof(detail::wtype)) {
shape.append(2);
d["typestr"] = python::str("<f") + python::str(sizeof(double));
} else {
d["typestr"] = python::str("<u") + python::str(self.depth());
}
d["shape"] = python::tuple(shape);
d["data"] = python::make_tuple(reinterpret_cast<boost::uintptr_t>(self.buffer()), false);
return d;
}
void simple_k_means( histogram<double>& hist, std::vector<double>& mu,int k_means,int maxiter)
{
std::vector<int> cls(hist.size(),0);
if(mu.size()!=k_means)
{
mu.resize(k_means);
//initializing k-means using uniform distribution
double vol_min=hist.find_percentile(0.001);
double vol_max=hist.find_percentile(0.999);
for(int j=0;j<k_means;j++)
mu[j]= vol_min+(vol_max-vol_min)*j/(double)(k_means-1);
}
for(int iter=0;iter<maxiter;iter++)
{
apply_k_means_classify(hist,mu,cls);
estimate_mu(hist,cls,mu);
}
}
// Run skin-detection algorithm
frame detect_skin(const frame& in) {
dtn_frame = in;
const rgb_byte ZERO = { };
for(int y = HEIGHT - 1; y; --y) {
for(int x = WIDTH - 1; x; --x) {
auto orig = dtn_frame.get_pixel(x, y);
rgb hist_in = { orig[2], orig[1], orig[0] };
if(hist.value(hist_in) < tld) {
orig[0] = orig[1] = orig[2] = 0;
}
}
}
return dtn_frame;
}
//! estimate sample mu using discrete classes
void estimate_mu(const histogram<double>& input,
const std::vector<int>& cls,
std::vector<double>& mu )
{
int number_of_classes=mu.size();
int i,j,k;
std::vector<double> _counts(number_of_classes,0);
for(k=0;k<number_of_classes;k++)
mu[k]=0;
double _count=0;
//1 calculate means
for(j=0;j<cls.size();j++)
{
_count+=input[j];
int _c=cls[j];
if(!_c || _c>number_of_classes) continue; //only use classified voxel
_c--;
_counts[_c]+=input[j];
mu[_c]+=input[j]*input.value(j);
}
if(!_count)
REPORT_ERROR("No voxels defined in ROI!");
for(k=0;k<number_of_classes;k++)
{
for(k=0;k<number_of_classes;k++)
{
if(_counts[k]>0)
mu[k]/=_counts[k];
}
}
}
// protected functions
void medianCut::computeBoxInfo(const histogram& hist,
boxInfo& theBox) const {
// boxInfo.min and .max must be specified before entry and histogram
// must be valid. Missing information in boxInfo is computed (mean,
// var, colorFrequency, colors) and box boundaries (min,max) are set
// to the smallest size, that still encloses all entries in the
// specified range of the histogram.
int rLow=theBox.min.getRed();
int gLow=theBox.min.getGreen();
int bLow=theBox.min.getBlue();
int rUp=theBox.max.getRed();
int gUp=theBox.max.getGreen();
int bUp=theBox.max.getBlue();
int rMin=rUp,rMax=rLow,gMin=gUp,gMax=gLow,bMin=bUp,bMax=bLow;
int i=rLow,j=gLow,k=bLow;
int freq;
ivector iVec(3);
double meanR, meanG, meanB, meanSquareR, meanSquareG, meanSquareB;
meanR = meanG = meanB = meanSquareR = meanSquareG = meanSquareB = 0;
double accu = 0;
theBox.colors = 0;
for (i=rLow;i<=rUp;i++) {
for (j=gLow;j<=gUp;j++) {
for (k=bLow;k<=bUp;k++) {
iVec[0]=i;
iVec[1]=j;
iVec[2]=k;
if (hist.at(iVec)>0.0f) {
if (k<bMin) {bMin=k;}
if (k>bMax) {bMax=k;}
if (j<gMin) {gMin=j;}
if (j>gMax) {gMax=j;}
if (i<rMin) {rMin=i;}
if (i>rMax) {rMax=i;}
freq = static_cast<const int>(hist.at(iVec));
meanR += freq * i;
meanG += freq * j;
meanB += freq * k;
meanSquareR += freq *i*i;
meanSquareG += freq *j*j;
meanSquareB += freq *k*k;
accu += freq;
// Count number of distinct colors
theBox.colors++;
}
}
}
}
meanR /= accu;
meanG /= accu;
meanB /= accu;
meanSquareR /= accu;
meanSquareG /= accu;
meanSquareB /= accu;
// Set minimum and maximum enclosing bounds
theBox.min.setRed(rMin);
theBox.min.setGreen(gMin);
theBox.min.setBlue(bMin);
theBox.max.setRed(rMax);
theBox.max.setGreen(gMax);
theBox.max.setBlue(bMax);
// Set number of entries inside box
theBox.colorFrequency = static_cast<long int>(accu);
// set mean and variance inside box
theBox.mean[0] = meanR;
theBox.mean[1] = meanG;
theBox.mean[2] = meanB;
theBox.var[0] = meanSquareR - meanR*meanR;
theBox.var[1] = meanSquareG - meanG*meanG;
theBox.var[2] = meanSquareB - meanB*meanB;
}
void sector_classifier::check_ngram_entropy()
{
/* Quickly compute the histogram */
/* Scan for a repeating ngram */
for(size_t ngram_size = 1; ngram_size < 20; ngram_size++){
bool ngram_match = true;
for(size_t i=ngram_size;i<sbuf.pagesize && ngram_match;i++){
if(sbuf[i%ngram_size]!=sbuf[i]) ngram_match = false;
}
if(ngram_match){
std::stringstream ss;
ss << CONSTANT << "(";
for(size_t i=0;i<ngram_size;i++){
char buf[16];
snprintf(buf,sizeof(buf),"0x%02x",sbuf[i]);
if(i>0) ss << ' ';
ss << buf;
}
ss << ")";
bulk_fr->write(sbuf.pos0,ss.str(),"");
return; // ngram is better than entropy
}
}
/* Couldn't find ngram; check entropy and FF00 counts...*/
size_t ff00_count=0;
h.add(sbuf.buf,sbuf.pagesize);
h.calc_distribution();
for(size_t i=0;i<sbuf.pagesize-1;i++){
if(sbuf[i]==0xff && sbuf[i+1]==0x00) ff00_count++;
}
if(sbuf.pagesize<=4096){ // we only tuned for 4096
if(ff00_count>2 && h.unique_counts()>=220){
bulk_fr->write(sbuf.pos0,JPEG,"");
return;
}
}
float entropy = h.entropy();
std::stringstream ss;
if(entropy>opt_high_entropy){
float cosineVariance = sd_autocorrelation_cosine_variance();
if(debug & DEBUG_INFO) ss << "high entropy ( S=" << entropy << ")" << " ACV= " << cosineVariance << " ";
if(cosineVariance > opt_MinimumCosineVariance){
ss << RANDOM;
} else {
ss << HUFFMAN;
}
bulk_fr->write(sbuf.pos0,ss.str(),"");
}
// don't bother recording 'low entropy' unless debuggin
if(entropy<=opt_high_entropy && opt_low_entropy) {
ss << "low entropy ( S=" << entropy << ")";
if(h.unique_counts() < 5){
ss << " Unique Counts: " << h.unique_counts() << " ";
for(std::vector<histogram::hist_element *>::const_iterator it = h.counts.begin();it!=h.counts.end();it++){
ss << (int)((*it)->val) << ":" << (*it)->count << " ";
}
}
}
}
