danila-lib 3.1.0

This is the module for detecting and classifying text on rama pictures

danila_lib v1.3.7

python library for Danila

To install project made

pip install danila-lib

To use in your project

from danila.danila import Danila

All use methods are in

class Danila

main method returns dict {'number', 'prod', 'year'} for openCV rama img or 'no_rama'

def text_recognize(self, img):

steps for algorythm

returns string - class of rama, img - openCV frame

def rama_classify(self, img):

returns openCV frame with rama from openCV frame

def rama_detect(self, img):

returns openCV image with cut_rama

def rama_cut(self, img):

returns openCV cut rama with drawn text areas

def text_detect_cut(self, img):

returns openCV img with drawn text areas

def text_detect(self, img):

in package data/neuro there is module Rama_classify_class

class Rama_classify_class

reads CNN taught model and includes it in class example

def __init__():

makes grey NumPy Array(1,512,512) of doubles[0..1] from openCV image

def prepare_img(img : openCV frame): NumPy Array(1,512,512)[0..1]

classify openCV img with CNN, returns list with double[0..1] values

def work_img(img : openCV frame): Double[0..1] list

classify openCV img with CNN, returns Class_im

def classify(img : openCV frame): Class_im

in package data/neuro there is module Rama_detect_class

class Rama_detect_class

reads yolov5 taught model from yandex-disk and includes it in class example

def __init__(self, model_path, model_name, yolo_path):

получить JSON с результатами yolo

def work_img(self, img_path):

получить координаты прямоугольника с рамой

def rama_detect(self, img_path):

in package data/neuro there is module Rama_text_detect_class

class Rama_text_detect_class

reads yolov5 taught model from yandex-disk and includes it in class example

def __init__(self, model_path, model_name, yolo_path):

find text areas on img from img_path with yolov5, returns yolojson

def work_img(self, img_path):

find text areas on img from img_path with yolov5, returns dict with rects for each text class

def text_detect(self, img_path):

draw img_text_areas on img, returns opencv img

def draw_text_areas_in_opencv(self, image_text_areas, img):

in package data/neuro there is module Letters_recognize

class Letters_recognize:

main_method takes all image_text_areas from image_rama_cut and recognize text

def work_image_cut(self, image_text_areas, image_rama_cut, number_length, prod_length, year_length):

read CNN model from yandex and put into object

def __init__(self):

cut text_areas imgs for each Rect from rect_array returns openCv imgs list

def make_cuts(self, img_rama_cut, rect_array):

for every text_class recognize text from all areas of text_class, length is depends on class and prod, returns string

def work_image_cuts(self, number_image_cuts, length):

recognize one word of given length from one img, returns str

def work_img_word(self, image_number, letter_number):

prepare img of one letter for CNN, returns np_array(1,28,28,1) of Double[0..1]

def prepare_img_letter(self, image_letter):

recognize img of one letter with CNN, returns list[10] of p

def work_img_letter(self, image_initial):

recognize img of one letter with CNN, returns letter in str

def classify_letter(self, image_letter):

in package data/result Rect module for rectangle operations

прочитать из json результата йоло

@staticmethod
def get_rect_from_yolo_json(yolo_json):

makes Rect object from xmin, xmax, ymin, ymax

def __init__(self, xmin=0, xmax=0, ymin=0, ymax=0):

Найти IOU между этим прямоугольником и другим, данным в объекте

def IoU(self, rect):

makes string from object

def __str__(self):

find intersection square between object and other rectangle

def intersection(self, rect):

find union RECT between object and other rectangle

def union(self, rect):

in package data/result Class_im

class Class_im(Enum):
    rama_no_spring = 0
    rama_spring = 1

in package data/result class Text_area

def __init__(self, dict_text_area):
    self.class_im = Class_text(dict_text_area['class'])
    self.rect = Rect(...)

in package data/result class image_text_areas

class contains dict with Rects list for each text_class

class Image_text_areas:

makes dict {Class_text.number : [], Class_text.prod : [], Class_text.text : [], Class_text.year : []}

def __init__(self):

add text area to dict

def add_area(self, text_area):

add list of text areas

def fill_in_with_areas(self, areas):

delete all cases in which two areas are intersected

def correct_intersections(self):

changes Rects coordinates from cut_img to whole_img from rama Rect

def explore_to_whole_image(self, rama_rect):

exapmles of using you can find

https://github.com/Arseniy-Zhuck/danila_lib_demo