First working version

.gitignore

@@ -0,0 +1,15 @@
+*.pyc
+.DS_Store
+*.swp
+dist/
+*.egg-info
+.project
+.pydevprojectthumbor/tests/visual_test/img
+.coverage
+tests/visual_test/img/
+*.swo
+*.swm
+*.swn
+tests/visual_test/build
+build
+*.so

remotecv/__init__.py

@@ -0,0 +1 @@
+__version__ = '0.1.0'

remotecv/app.py

@@ -0,0 +1,13 @@
+import bson
+
+from remotecv.image_processor import ImageProcessor
+
+class RemoteCvApp:
+
+    def __init__(self):
+        self.processor = ImageProcessor()
+
+    def process_request(self, msg):
+        msg = bson.loads(msg)
+        points = self.processor.detect(msg['type'], msg['size'], msg['mode'], msg['image'])
+        return bson.dumps({ 'points': points })

remotecv/detectors/__init__.py

@@ -0,0 +1,80 @@
+#!/usr/bin/python
+# -*- coding: utf-8 -*-
+
+# thumbor imaging service
+# https://github.com/globocom/thumbor/wiki
+
+# Licensed under the MIT license:
+# http://www.opensource.org/licenses/mit-license
+# Copyright (c) 2011 globo.com timehome@corp.globo.com
+
+from os.path import join, dirname, abspath, isabs
+
+import cv
+
+class BaseDetector(object):
+
+    def detect(self, context):
+        raise NotImplementedError()
+
+
+class CascadeLoaderDetector(BaseDetector):
+
+    def load_cascade_file(self, module_path, cascade_file_path):
+        if not hasattr(self.__class__, 'cascade'):
+            if isabs(cascade_file_path):
+                cascade_file = cascade_file_path
+            else:
+                cascade_file = join(abspath(dirname(module_path)), cascade_file_path)
+            setattr(self.__class__, 'cascade', cv.Load(cascade_file))
+
+    def get_min_size_for(self, size):
+        ratio = int(min(size[0], size[1]) / 15)
+        ratio = max(20, ratio)
+        return (ratio, ratio)
+
+    def get_features(self, width, height, mode, img_data):
+        sz = (width, height)
+        image = cv.CreateImageHeader(sz, cv.IPL_DEPTH_8U, 3)
+        cv.SetData(image, img_data)
+
+        gray = cv.CreateImage(sz, 8, 1);
+        convert_mode = getattr(cv, 'CV_%s2GRAY' % mode)
+        cv.CvtColor(image, gray, convert_mode)
+
+        # min_size = (20, 20)
+        min_size = self.get_min_size_for(sz)
+        haar_scale = 1.2
+        min_neighbors = 1
+
+        cv.EqualizeHist(gray, gray)
+
+        faces = cv.HaarDetectObjects(gray,
+                                     self.__class__.cascade, cv.CreateMemStorage(0),
+                                     haar_scale, min_neighbors,
+                                     cv.CV_HAAR_DO_CANNY_PRUNING, min_size)
+
+        faces_scaled = []
+
+        for ((x, y, w, h), n) in faces:
+            # the input to cv.HaarDetectObjects was resized, so scale the
+            # bounding box of each face and convert it to two CvPoints
+            pt1 = (x, y)
+            pt2 = ((x + w), (y + h))
+            x1 = pt1[0]
+            x2 = pt2[0]
+            y1 = pt1[1]
+            y2 = pt2[1]
+            faces_scaled.append(((x1, y1, x2-x1, y2-y1), None))
+
+        return faces_scaled
+
+    def detect(self, context):
+        features = self.get_features(width, height, mode, img_data)
+
+        if features:
+            points = [[left, top, width, height] for (left, top, width, height), neighbors in features]
+        else:
+            points = []
+
+        return points

remotecv/detectors/face_detector/__init__.py

@@ -0,0 +1,33 @@
+#!/usr/bin/python
+# -*- coding: utf-8 -*-
+
+# thumbor imaging service
+# https://github.com/globocom/thumbor/wiki
+
+# Licensed under the MIT license: 
+# http://www.opensource.org/licenses/mit-license
+# Copyright (c) 2011 globo.com timehome@corp.globo.com
+
+from detectors import CascadeLoaderDetector
+
+HAIR_OFFSET = 0.12
+
+class FaceDetector(CascadeLoaderDetector):
+
+    def __init__(self):
+        self.load_cascade_file(__file__, 'haarcascade_frontalface_alt.xml')
+
+    def __add_hair_offset(self, top, height):
+        top = max(0, top - height * HAIR_OFFSET)
+        return top
+
+    def detect(self, width, height, mode, img_data):
+        features = self.get_features(width, height, mode, img_data)
+
+        points = []
+        if features:
+            for (left, top, width, height), neighbors in features:
+                top = self.__add_hair_offset(top, height)
+                points.append([left, top, width, height])
+
+        return points