def _find_elements_by_mousehover(self, tab_screens=None, ui_changes=None): ''' Find elements in the active window using the tab key and mouse hovering techniques. ''' self._user_automation.mouse.move(1, 1) if tab_screens is None: tab_screens = get_tab_changing_areas(self._grab_screen, self._send_tab) if ui_changes is None: ui_changes = get_ui_changes(tab_screens) start_screen = Image2(self._grab_screen()) extracted_areas = [] points = get_corner_points(ui_changes, 3, 3) for point in points: if not point_inside_rects((point[0], point[1]), extracted_areas): self._user_automation.mouse.move(point[0], point[1]) current_screen = Image2(self._grab_screen()) if current_screen != start_screen: changed_area = current_screen.difference(start_screen) if not changed_area in extracted_areas: if point_inside_rects((point[0], point[1]), [changed_area]): extracted_areas.append(changed_area) return extracted_areas
def test(): ''' Simple unit test ''' img1 = Image2(file_name='self_test/img/active_window.bmp') img2 = Image2(file_name='self_test/img/inactive_window.bmp') diff = crop_border_differences(img1, img2) crop_1 = img1.image.crop((diff[0], diff[1], diff[2] + 1, diff[3] + 1)) crop_1.save('crop1.bmp') crop_2 = img2.image.crop((diff[0], diff[1], diff[2] + 1, diff[3] + 1)) crop_2.save('crop2.bmp')
def get_verb_image_b64(model, verb): if type(verb['how']) is dict and len(verb['how']['snapshots']) > 0: how = verb['how']['snapshots'][0] image1 = Image2(file_name="%s/%s" % (model.images_dir, how)) else: image1 = Image.open( os.path.dirname(__file__) + "/noimageavailable.png") image1 = Image2(image=image1) image1.image.save("tmp.png") return base64.b64encode(read_binary_file("tmp.png"))
def parametrize(self, ignorable_areas): ''' For the given node we need to mark as dynamic the changing areas and ensure recognition of it will go smooth FIXME: should check sanity validate on tab images ''' if self.file_name is None: raise Exception("Node file name must be set before parametrize!") if len(self.reference_images) == 0: raise Exception("No reference images available to parametrize") dialog = self.reference_images[0].image color = (237, 28, 36, 0) for edge in self.edges: if edge.location: elem_rect = expand_rect(edge.location, (1, 1)) dialog.paste(color, elem_rect) LOGGER.debug("Parametrizing with %s ignorable areas" % len(ignorable_areas)) for area in ignorable_areas: elem_rect = expand_rect(area, (1, 1)) dialog.paste(color, elem_rect) dialog = Image2(image=dialog, color_mask=0xed1c2400, tolerance=self.graph.image_tolerance) self.screenshots.append(dialog)
def get_node_image_b64(model, view, reference_image=False): if not reference_image: view_images = view['self'].HERE.get('snapshots', []) else: view_images = view['self'].HERE.get('reference snapshots', []) image1 = Image2(file_name="%s/%s" % (model.images_dir, view_images[0])) image1.image.save("tmp.png") return base64.b64encode(read_binary_file("tmp.png"))
def perform(self, world): LOGGER.info("Performing %s" % str(self)) screen = world.machine.automation.grab_screen() world.last_screen = screen.copy() if self.method: self.method(world) # pylint: disable=E1102 else: #found = self.tail.ui_element.find_in(Image2(image=screen)) found = self.tail.find_in(Image2(image=screen)) if found: #links usually need to be clicked close to the left #top coord (if left aligned) but some buttons needs #the center as they shade in the borders if self.ui_type == 'link': x_coord = found[0] + self.location[0] + 4 y_coord = found[1] + self.location[1] + 4 else: center_x, center_y = center_of_rect(self.location) x_coord = found[0] + center_x y_coord = found[1] + center_y #Some quite seldom and random issues when clicking!? world.machine.automation.mouse.move(x_coord - 1, y_coord - 1) world.machine.automation.mouse.click(x_coord, y_coord) if self.ui_type == 'desktop icon': world.machine.automation.mouse.click() elif self.ui_type == 'text' and 'test value' in self.custom: world.machine.automation.keyboard.enters( self.custom['test value'][0]['value']) else: screen.save("Current screen.bmp") raise Exception("Parent node (%s) not found in screen!" % self.tail.name) #FIXME: for recording purposes this is needed, for playback purposes #the destination should be checked by waiting... world.machine.automation.mouse.move(1, 1) world.machine.wait_idling() if self.head: destination_reached = False # FIXME: this is not entirely correct, some long transitions may # happen for example during installations or parts of the # applications were it looks like stuck for a while for i in range(3): destination_reached = self.head.is_in(world) if destination_reached: break LOGGER.info("Expected destination not in screen yet, " "waiting (%s)" % i) world.machine.wait_idling() if destination_reached == False: LOGGER.info("Expected destination not in screen, could be a " "type of return to caller scenario") elif self.head.enter_hook: self.head.enter_hook(world)
def import_from_file(self, file_name, images_dir): ''' Imports this node from a file, images_dir does not replace the node directory or images_dir ''' with open(file_name, "rb") as the_file: json_node = json.load(the_file) self.name = json_node['HERE']['desc'] use_name = file_name if use_name.find('.') != -1: use_name = use_name.split('.')[0] self.file_name = use_name.split("/")[-1].split("\\")[-1] for reference in json_node['HERE']['reference screenshots']: an_image = Image2(file_name="%s/%s" % (images_dir, reference['file']), color_mask=reference.get('mask')) an_image.file_name = None self.reference_images.append(an_image) for screenshot in json_node['HERE']['screenshots']: an_image = Image2(file_name="%s/%s" % (images_dir, screenshot['file']), color_mask=screenshot.get('mask')) an_image.file_name = None self.screenshots.append(an_image) self.title = json_node['HERE'].get('title', '') keys = list(json_node.keys()) keys.remove('HERE') keys.sort() for i in range(len(keys)): elem = json_node[keys[i]] edge = self.create_edge(elem['desc']) #FIXME: implement edge.head set edge.location = elem['how']['visual'] edge.ui_type = elem['how']['type'] for screenshot in elem['how']['screenshots']: an_image = Image2(file_name="%s/%s" % (images_dir, screenshot['file']), color_mask=screenshot.get('mask')) an_image.file_name = None edge.screenshots.append(an_image)
def get_tab_changing_areas(grab_screen, send_tab): ''' Returns an array of images produced by cycling with tab over a window Put the mouse out of the way before calling this cases: cycles thru controls, looping ok there's only 1 control with focus already there's only 1 control, focus not set (2 screenshots) one of the controls will 'capture' and wont let the focus go out ''' last_screen = Image2(grab_screen()) images = [last_screen] tail = 0 repeated = 0 while True: last_screen = send_tab_expecting_change(send_tab, grab_screen, last_screen) images.append(last_screen) #consecutive match? if images[tail] == images[-1]: tail += 1 repeated += 1 if repeated > 2: break else: tail = 0 repeated = 0 #search if it starts repeating somewhere not in 1st screenshot for position in range(len(images) - 1): if images[position] == images[-1]: tail = position + 1 repeated = 1 break if len(images) > 30: LOGGER.warning("Stopped tab cycling after 30 different " "screenshots...") break for _ in range(repeated): images.pop() #remove repeated images in the tail if there are some while len(images) > 1 and images[-1] == images[-2]: images.pop() for _ in range(repeated - 1): send_tab(backwards=True) LOGGER.debug("Got %s tab images" % len(images)) return images
def __init__(self, user_simulation=None, crop_taskbar=True): if not user_simulation: from murphy.user_simulation.local.local_user import User self._user_automation = User() else: self._user_automation = user_simulation path = os.path.dirname(__file__) + '/img' radio_images = [ Image2(file_name=path + '/radio1.bmp', color_mask=0xed1c2400), Image2(file_name=path + '/radio2.bmp', color_mask=0xed1c2400) ] self._radio = UIElement(radio_images) check_images = [ Image2(file_name=path + '/check1.bmp', color_mask=0xed1c2400), Image2(file_name=path + '/check2.bmp', color_mask=0xed1c2400) ] self._checkboxes = UIElement(check_images) self._crop_taskbar = crop_taskbar self._internal_counter = 0
def compare_view_images(self, model, view, reference_model, reference_view): #we compare parametrized images, otherwise dates and times will not match view_images = view['self'].HERE.get('snapshots', []) reference_view_images = reference_view['self'].HERE.get( 'snapshots', []) if len(view_images) != len(reference_view_images): return 'Node "%s" has %s snapshots but the reference node has %s' % ( view['self'].HERE['desc'], len(view_images), len(reference_view_images)) result = '' for index in range(len(view_images)): image1 = Image2(file_name="%s/%s" % (model.images_dir, view_images[index])) image2 = Image2( file_name="%s/%s" % (reference_model.images_dir, reference_view_images[index]), tolerance=0.9999) if image1 != image2: result += 'Node "%s", image "%s" differs from reference node "%s"<br>\n' % ( view['self'].HERE['desc'], view_images[index], reference_view_images[index]) image1.image.save("tmp.png") encoded1 = base64.b64encode(read_binary_file("tmp.png")) image2.image.save("tmp.png") encoded2 = base64.b64encode(read_binary_file("tmp.png")) id1 = str(uuid.uuid1()) id2 = str(uuid.uuid1()) result += "<table>\n\t<tr>\n\t\t<td><img id='%s' title='model' src='data:image/png;base64,%s'></td>\n" % ( id1, encoded1) result += "\t\t<td><input style='font-family:\"Courier New\", Courier, monospace;' type=button id='button-%s' value='<- New \n Reference ->' onclick='swap(\"%s\", \"%s\")'></td>\n" % ( id1, id1, id2) result += "\t\t<td><img id='%s' title='reference' src='data:image/png;base64,%s'></td>\n\t</tr>\n</table><br>\n" % ( id2, encoded2) return result
def get_elements_by_hover_points(self, window_origin, hover_points): ''' Simple recognition based on hovering the mouse thru the given list of points, the points should be close to the center of the elements and the elements must highlight on mouseover to work properly Points are relative to the window, not absolute screen coords ''' self._user_automation.mouse.move(1, 1) screen = Image2(self._grab_screen()) result = [] for point in hover_points: hover_x = window_origin[0] + point[0] hover_y = window_origin[1] + point[1] self._user_automation.mouse.move(hover_x, hover_y) self._user_automation.mouse.move(hover_x + 1, hover_y + 1) screen_when_hover = Image2(self._grab_screen()) coords = screen.difference(screen_when_hover) if coords: cursor = self._user_automation.get_current_cursor() element = {'coords': coords, 'type': cursor} result.append(element) return result
def wait(self, element, max_wait=60, retry_every=0.3): ''' Waits for the given UI element to be visible, returns the bounding coordinates of the elements if found, raises ValueError if not. ''' started_at = datetime.datetime.now() while True: screen = Image2(image=self.grab_screen()) result = element.find_in(screen) if result: return result now = datetime.datetime.now() if (now - started_at).seconds > max_wait: raise ValueError("Element not found in screen") time.sleep(retry_every)
def send_tab_expecting_change(send_tab, grab_screen, current_screen): ''' Sends a tab key to change control focus, wait few seconds for any ui change to synchronize it, will give up after 5 seconds (because sometimes focus is not properly handled in applications) Returns an Image2 of the new screen ''' send_tab() for _ in range(5): new_screen = Image2(grab_screen()) if new_screen != current_screen: break time.sleep(1) return new_screen
def is_in(self, world): ''' Checks if this node is recognizable at this moment in the screen, returns either True or False ''' LOGGER.debug("Checking if i'm in %s" % self.name) if ((len(self.edges) == 1 and self.edges[0].ui_type == 'desktop icon') or self.desktop_icon == True): rect = world.machine.helper.get_active_window_rect() if rect and rect[0] > 0 and rect[1] > 0: LOGGER.debug("Not there, desktop icons are detectable when " "there are no active windows") return False screen = Image2(image=world.machine.automation.grab_screen()) found = self.find_in(screen) if found: LOGGER.debug("Found") return True else: LOGGER.debug("Not found") return False
def compare(self): views = self.model.new_worker().get_views() reference_views = self.reference_model.new_worker().get_views() matching_views = [] candidate_for_moved = [] moved_views = [] candidate_for_new = [] changed_views = [] new_views = [] reference_views_used = [] #dictionary of moved nodes, key is model view nam, value is reference view name reference_translation = {} reference_translation[''] = '' result = '' ordered_views = self._get_ordered_views(self.model) for view_name in ordered_views: view = views[view_name] if view_name in reference_views: reference_view = reference_views[view_name] comparison = self.compare_view_images(self.model, view, self.reference_model, reference_view) if comparison == '': matching_views.append(view) reference_views_used.append(reference_view) reference_translation[view_name] = view_name else: candidate_for_moved.append(view) else: candidate_for_moved.append(view) for view in candidate_for_moved: candidate = self.find_node_by_image(self.model, view, self.reference_model, reference_views) if candidate and not candidate in reference_views_used: moved_views.append({'view': view, 'reference view': candidate}) reference_views_used.append(candidate) reference_translation[ view['self'].HERE['desc']] = candidate['self'].HERE['desc'] else: candidate_for_new.append(view) for view in candidate_for_new: view_name = view['self'].HERE['desc'] if view_name in reference_views and not reference_views[ view_name] in reference_views_used: #can this still be wrong? changed_views.append(view) reference_views_used.append(reference_views[view_name]) reference_translation[view_name] = view_name else: new_views.append(view) #print "Translation table: %s" % str(reference_translation) for view in matching_views: reference_view = reference_views[view['self'].HERE['desc']] result += self.compare_edges(view, reference_view, reference_translation) for movement in moved_views: result += "Node '%s' is in reference model as %s<br>\n" % ( movement['view']['self'].HERE['desc'], movement['reference view']['self'].HERE['desc']) result += self.compare_edges(movement['view'], movement['reference view'], reference_translation) for view in changed_views: result += self.compare_view_images( self.model, view, self.reference_model, reference_views[view['self'].HERE['desc']]) for view in new_views: result += "Node '%s' is new, does not seems to exist in the reference model<br>\n" % ( view['self'].HERE['desc']) view_images = view['self'].HERE.get('snapshots', []) if len(view_images) > 0: image = Image2(file_name="%s/%s" % (self.model.images_dir, view_images[0])) image.image.save("tmp.png") encoded = base64.b64encode(read_binary_file("tmp.png")) result += "<img src='data:image/png;base64,%s' title='model'><br>\n" % encoded for view in reference_views.values(): if not view in reference_views_used: result += "Reference node '%s' does not seem to exists in the model (was removed?)<br>\n" % ( view['self'].HERE['desc']) view_images = view['self'].HERE.get('snapshots', []) if len(view_images) > 0: image = Image2( file_name="%s/%s" % (self.reference_model.images_dir, view_images[0])) image.image.save("tmp.png") encoded = base64.b64encode(read_binary_file("tmp.png")) result += "<img src='data:image/png;base64,%s' title='model'><br>\n" % encoded return result
'''
def scrap_state(node, world, scraper_hints, node_hints): ''' Scraps the ui into the given node for the given world state ''' node_index = node.graph.nodes.index(node) node.name = 'Node %s' % node_index node.file_name = 'node_%s' % str(node_index).zfill(2) screen_image = world.machine.automation.grab_screen() screen = Image2(image=screen_image) LOGGER.debug("Node hints for scrapper, node %s are %s" % (node_index, str(scraper_hints))) is_desktop = False if scraper_hints.get('windowless', False): #rect is difference between world.last_screen and screen no_taskbar = world.last_screen no_taskbar = world.last_screen.crop( (0, 0, no_taskbar.size[0], no_taskbar.size[1] - TASKBAR_HEIGHT)) window_rect = Image2(image=no_taskbar).difference(screen) window_rect = automation_helpers.refine_window_rect( screen_image, window_rect) else: is_desktop, window_rect = solve_active_window_rect(world, screen_image) window_image = Image2(image=screen_image.crop(window_rect)) node.reference_images.append(window_image) node.last_location = window_rect #Don't scrap the desktop! print "Screen size %s, window_rect %s" % (str( screen_image.size), str(window_rect)) if is_desktop == "Screen": LOGGER.info( "Disabling outfocus method as window rect is the whole screen") scraper_hints['outfocus method'] = False is_desktop = False if not is_desktop: scraper = Scraper(world.machine.automation, crop_taskbar=True) if 'defined by hover points' in scraper_hints: LOGGER.debug("Node defined by hover points") elements = scraper.get_elements_by_hover_points( (window_rect[0], window_rect[1]), scraper_hints['defined by hover points']) else: elements = scraper.get_elements(scraper_hints, (window_rect[0], window_rect[1])) else: #FIXME: could search for new icons... elements = [] LOGGER.debug("Found elements: %s" % str(elements)) LOGGER.debug("Node hints for node %s are %s" % (node_index, str(node_hints))) ignorable_areas = node_hints.get('ignorable', []) for element in elements: element_rect = move_rect(element['coords'], (window_rect[0], window_rect[1])) if not is_rect_inside(element_rect, ignorable_areas): edge_name = 'Element %s' % len(node.edges) edge = node.create_edge(edge_name) edge.location = element_rect edge.absolute_location = element['coords'] edge.ui_type = element['type'] elem_rect = expand_rect(element['coords'], (1, 1)) screenshot = Image2(image=screen.image.crop(elem_rect)) edge.screenshots.append(screenshot) if edge.ui_type == 'text': edge.head = node else: LOGGER.debug( ("Ignoring element at %s as it is inside ignorable" + " area") % str(element_rect))
def get_elements(self, hints=None, window_origin=None): ''' Returns a list of UI elements that this class can identify from the currently active window. The return of this implementation is an array of dictionaries where each dictionary describes the control type and it's bounding box ''' if hints is None: hints = {} if window_origin is None: window_origin = (0, 0) if hints.get('outfocus method', False): result = self.get_elements_by_app_outfocus(hints, window_origin) if result != False: return result else: LOGGER.warning( 'Unable to properly use the outfocus hint, ' 'reverting to standard behaviour for this node.') self._user_automation.mouse.move(1, 1) screen = self._grab_screen() screen_height = screen.size[1] screen = Image2(screen) tab_screens = get_tab_changing_areas(self._grab_screen, self._send_tab) if len(tab_screens) == 1: LOGGER.info('Only one image obtained when cycling with tab, adding' ' alt trick.') self._user_automation.keyboard.enters('{alt}') #we're searching for very small clue here... just one _ new_screen = Image2(self._grab_screen(), tolerance=1.0) tab_screens.append(new_screen) candidates = [] processed = [] for i in range(len(tab_screens) - 1): coords = tab_screens[i].difference(tab_screens[i + 1]) if coords: LOGGER.debug("Changes from %s to %s are: %s" % (i, i + 1, str(coords))) division = automation_helpers.find_inner_bbox( tab_screens[i], tab_screens[i + 1], coords) LOGGER.debug("Splitting found %s bboxes (%s)" % (len(division), str(division))) for rect in division: if not rect in candidates: LOGGER.debug("Adding: %s" % str(rect)) candidates.append(rect) #hover, if image differs take diff coords, use biggest # of two use mouse pointer clue for type #ARGGGGGG cursor blinking... deactivated at os level for #now #the focus may be at this point anywhere and on 1st #case is where it is left from tab navigation, for #cases like menu we have to highlight current menu item center = center_of_rect(rect) self._user_automation.mouse.move(center[0], center[1]) self._user_automation.mouse.move( center[0] + 1, center[1] + 1) cursor = self._user_automation.get_current_cursor() screen1 = Image2(self._grab_screen()) self._user_automation.mouse.move(1, screen_height) screen2 = Image2(self._grab_screen()) diff = screen1.difference(screen2) if diff: #produced a change in UI, must be button LOGGER.debug(("Will compute biggest rect out of " "%s %s") % (str(rect), str(diff))) biggest_rect = get_bounding_box([rect, diff]) if not biggest_rect in processed: processed.append(biggest_rect) LOGGER.debug("Added: %s" % str(biggest_rect)) else: #no UI change, can be a link, text or misfired #recognition, exceptional case is one button alone #in dialog if ((cursor != 'normal' and not rect in processed) or (len(tab_screens) == 2 and not rect in processed)): processed.append(rect) LOGGER.debug("Added: %s" % str(rect)) LOGGER.debug("There are %s elements to consider from tab + hovering" % len(processed)) checkboxes = self._checkboxes.find_all(screen) LOGGER.debug("Found %s checkboxes" % len(checkboxes)) checkboxes = add_text_to_elements(checkboxes, screen) radios = self._radio.find_all(screen) LOGGER.debug("Found %s radios" % len(checkboxes)) radios = add_text_to_elements(radios, screen) checkboxes = merge_overlapping_areas(checkboxes, processed) radios = merge_overlapping_areas(radios, processed) areas = exclude_subareas(processed, checkboxes + radios) points = hints.get('points of interest', []) LOGGER.debug("Points of interest are: %s" % str(points)) for point in points: point_x = window_origin[0] + point[0] point_y = window_origin[1] + point[1] found, bbox = find_bounding_box(screen.image, point_x, point_y) if found: LOGGER.debug("Found %s from point of interest" % str(bbox)) areas.append(bbox) else: LOGGER.debug("Nothing found from point of interest at %s %s" % (point_x, point_y)) result = [] for area in areas: center_x, center_y = center_of_rect(area) self._user_automation.mouse.move(center_x, center_y) self._user_automation.mouse.move(center_x + 1, center_y + 1) cursor = self._user_automation.get_current_cursor() element = { 'coords': (area[0], area[1], area[2], area[3]), 'type': cursor } result.append(element) for area in checkboxes: element = { 'coords': (area[0], area[1], area[2], area[3]), 'type': 'checkbox' } result.append(element) for area in radios: element = { 'coords': (area[0], area[1], area[2], area[3]), 'type': 'radio' } result.append(element) result = remove_containers(result) return result
def get_elements_by_app_outfocus(self, hints=None, window_origin=None): ''' Alternate algorithm, tries to find the elements by setting the focus in the desktop and then back to the application while analyzing the difference on the screen, when the application loses the focus the normal behaviour in windows is that the active control is not shown with the focus This technique has some caveats, for example the border of the application needs to be ignored, also the default button will also be rendered differently giving in some cases 2 areas with changes ''' self._user_automation.mouse.move(1, 1) if hints is None: hints = {} if window_origin is None: window_origin = (0, 0) rects = [] repeated = 0 attempts = 0 while True: screen = Image2(self._grab_screen()) width = screen.size[0] self._user_automation.mouse.click(width / 2, screen.size[1] + 5) time.sleep(0.1) screen2 = Image2(self._grab_screen()) self._user_automation.keyboard.enters("{+alt}{tab}{+shift}" "{tab}{-shift}{-alt}") time.sleep(1) #coords = screen.difference(screen2) coords = automation_helpers.crop_border_differences( screen, screen2) crop1 = screen.image.crop( (coords[0], coords[1], coords[2] + 1, coords[3] + 1)) crop2 = screen2.image.crop( (coords[0], coords[1], coords[2] + 1, coords[3] + 1)) coords2 = Image2(image=crop1).difference(Image2(image=crop2)) if coords2: coords = (coords[0] + coords2[0], coords[1] + coords2[1], coords[0] + coords2[2], coords[1] + coords2[3]) divisions = automation_helpers.find_inner_bbox( screen, screen2, coords) LOGGER.debug("Splitting found %s bboxes (%s)" % (len(divisions), str(divisions))) repeated_in_divisions = 0 for rect in divisions: print "resulting coords %s" % str(rect) if rect in rects: repeated_in_divisions += 1 else: rects.append(rect) if repeated_in_divisions == len(divisions): repeated += 1 if repeated == 2: break self._user_automation.keyboard.enters("{tab}") time.sleep(0.1) attempts += 1 if attempts - len(rects) > 5: #something is wrong, is possible that is unable to detect #horizontal division at all return False result = [] for rect in rects: self._user_automation.mouse.move(1, 1) time.sleep(0.2) before_cursor = self._user_automation.get_current_cursor() before_screen = Image2(self._grab_screen()) center_x, center_y = center_of_rect(rect) self._user_automation.mouse.move(center_x, center_y) self._user_automation.mouse.move(center_x + 1, center_y + 1) time.sleep(0.2) after_screen = Image2(self._grab_screen()) after_cursor = self._user_automation.get_current_cursor() ui_changes = before_screen != after_screen cursor_changes = before_cursor != after_cursor everything = not hints.get('visual clue needed', True) if cursor_changes or ui_changes or everything: element = { 'coords': (rect[0], rect[1], rect[2], rect[3]), 'type': after_cursor } result.append(element) return result
def find_inner_bbox(image1, image2, bbox): ''' Tries to separate the changes inside bbox, returns an array with either 1 or 2 bboxes ''' result = [] division = find_vertical_division(image1, image2, bbox) if division: print "found vertical division as %s" % (str(division)) a_box = [bbox[0], bbox[1], bbox[2]+1, bbox[3]+1] img1 = Image.new("RGB", image1.size) img1.paste(image1.image.crop(a_box), (bbox[0], bbox[1])) img2 = Image.new("RGB", image2.size) img2.paste(image2.image.crop(a_box), (bbox[0], bbox[1])) img3 = img1.copy() img3.paste((0, 0, 0), [division[1], 0, img1.size[0], img1.size[1]]) img4 = img2.copy() img4.paste((0, 0, 0), [division[1], 0, img2.size[0], img2.size[1]]) bbox1 = Image2(image=img3).difference(Image2(image=img4)) if bbox1: result.append(bbox1) img3 = img1.copy() img3.paste((0, 0, 0), [0, 0, division[0] + 1, img1.size[1]]) img4 = img2.copy() img4.paste((0, 0, 0), [0, 0, division[0] + 1, img2.size[1]]) bbox2 = Image2(image=img3).difference(Image2(image=img4)) if bbox2: result.append(bbox2) if len(result) == 0: result.append(bbox) return result division = find_horizontal_division(image1, image2, bbox) if division: print "found horizontal division" a_box = [bbox[0], bbox[1], bbox[2]+1, bbox[3]+1] img1 = Image.new("RGB", image1.size) img1.paste(image1.image.crop(a_box), (bbox[0], bbox[1])) img2 = Image.new("RGB", image2.size) img2.paste(image2.image.crop(a_box), (bbox[0], bbox[1])) img3 = img1.copy() img3.paste((0, 0, 0), [0, division[1], img1.size[0], img1.size[1]]) img4 = img2.copy() img4.paste((0, 0, 0), [0, division[1], img2.size[0], img2.size[1]]) bbox1 = Image2(image=img3).difference(Image2(image=img4)) if bbox1: result.append(bbox1) img3 = img1.copy() img3.paste((0, 0, 0), [0, 0, img1.size[0], division[0] + 1]) img4 = img2.copy() img4.paste((0, 0, 0), [0, 0, img2.size[0], division[0] + 1]) bbox2 = Image2(image=img3).difference(Image2(image=img4)) if bbox2: result.append(bbox2) if len(result) == 0: result.append(bbox) return result return [bbox]