image-stitcher

Python library for general mass image stitching.

Setup

pip install image-stitcher

Usage

An example can be found in tests/.

Arbitrary stitching

Relations in an arbitrarily arranged set of images are automatically determined. Each related collection found is stitched together.

import os
from image_stitcher.stitcher import ArbitraryStitcher

dir_path = "path/to/dir"
filepaths = [os.path.join(dir_path, f) for f in os.listdir(dir_path)]

arbs = ArbitraryStitcher(
    log_level = 2,
    output_dir = "stitches/",
    min_match_count = 5,
    lowes_ratio_threshold = 0.65,
    algorithm = 0,
    matcher = 0,
    type = 0,
    resize = None,
    consecutive_range = None,
    blend_processor = lambda x: x,
    blender = 1,
    ref_image_contrib = 0.2,
    images = None, 
    filepaths = dir_path, 
    with_tqdm = True
)

collections = arbs.get_collections()
stitched = arbs.stitch_collections(collections)

Consecutive stitching

• Images are sequentially stitched without pre-determining relations in the large set.
• Optimizations coded in allow this to stitch HD images at 5Hz and SD images at up to 10Hz. Suitable for applications such as live aerial drone mapping.

import os
from image_stitcher.stitcher import ConsecutiveStitcher


dir_path = "path/to/dir"
filepaths = [os.path.join(dir_path, f) for f in os.listdir(dir_path)]


def load_image(filepath, to_resize = None):
    image = cv2.imread(filepath)
    return resize_image(image, to_resize)

cons = ConsecutiveStitcher(
    log_level = 2,
    output_dir = "stitches/",
    min_match_count = 5,
    lowes_ratio_threshold = 0.65,
    algorithm = 0,
    matcher = 0,
    type = 0,
    resize = None,
    consecutive_range = None,
    blend_processor = lambda x: x,
    blender = 1,
    ref_image_contrib = 0.2,
    consecutive_range = 1,
    backup_interval = False,
    consecutive_volatility_threshold = 4,
)

for filepath in filepaths:
    cons.stitch_consecutive(load_image(filepath))

Synchronous stitching

Consecutively stitches a second set of images using the same transformations calculated when stitching the first set. Can be used to stitch depth images from corresponding RGB images.

from image_stitcher.stitcher import ConsecutiveStitcher


class SynchronizedStitcher:
    def __init__(self, rgb_args=None, depth_args=None, points=None):
        self.rgb_stitcher = ConsecutiveStitcher(**rgb_args)
        self.depth_stitcher = ConsecutiveStitcher(**depth_args)

    def stitch(self, rgb_image, depth_image = None):
        first_stitch = self.rgb_stitcher.stitch_count == 0
        tf = self.rgb_stitcher.stitch_consecutive(rgb_image)
        if not (depth_image is None):
            if not first_stitch:
                if self.rgb_stitcher.stitch_count > self.depth_stitcher.stitch_count:
                    self.depth_stitcher.save_and_reset()
            self.depth_stitcher.stitch_consecutive(depth_image, tf)
        return tf


sync_st = SynchronizedStitcher(
    rgb_args={
        "output_dir": "path/to/rgbdir",
        "matcher": 1,
        "algorithm": 1,
        "backup_interval": 50,
        "consecutive_range": 2,
        "blender": 1,
        "log_level": 2
    },
    depth_args={
        "output_dir": "path/to/depthdir",
        # "blend_processor": lambda x: utils.color_map(x),
        "backup_interval": 50,
        "blender": 0,
        "ref_image_contrib": 0.5,
        "log_level": 2
    }
)