--- a
+++ b/FastRCNN/utils/imageUtils.py
@@ -0,0 +1,70 @@
+import matplotlib.pyplot as plt
+import numpy as np
+
+
+def cropImage(img, bboxes, expand=True):
+    """crop images by the given bounding boxes.
+
+    Args:
+        img (numpy.ndarray): image in CHW format
+        bboxes (numpy.ndarray): bounding boxes in the format specified by chainerCV
+        expand (bool): whether to expand the bounding boxes or not
+
+    Returns:
+        a batch of cropped image in CHW format
+        The image is in CHW format and its color channel is ordered in
+        RGB.
+
+    Return type: list
+
+    """
+
+    if expand:
+        _, H, W = img.shape
+        bboxes = expand_bbox(bboxes, H, W)
+
+    subimages = list()
+    for bbox in bboxes:
+        bbox = bbox.astype(np.int)
+        subimages.append(img[:, bbox[0]:bbox[2]+1, bbox[1]:bbox[3]+1])
+    return subimages, bboxes
+
+
+def expand_bbox(bbox, H, W):
+    """
+    expand the bounding box within the range of height and width of the image
+    :param bbox: numpy.ndarray bounding box N by 4
+    :param H: int Height of the image
+    :param W: int Width of the image
+    :return: numpy.ndarray expanded bounding box
+    """
+    b_height = 0.15*(bbox[:, 2] - bbox[:, 0])
+    b_width = 0.15*(bbox[:, 3] - bbox[:, 1])
+    b_height[b_height < 7] = 7
+    b_width[b_width < 7] = 7
+    adjust = np.array((-b_height, -b_width, b_height, b_width)).transpose()
+    new_bbox = bbox + adjust
+    new_bbox[new_bbox < 0] = 0
+    new_bbox[new_bbox[:, 2] >= H, 2] = H - 1
+    new_bbox[new_bbox[:, 3] >= W, 3] = W - 1
+
+    return new_bbox
+
+
+def showImage(img):
+    """
+    :param img (numpy.ndarray): image in CHW format
+    :return: plot the red channel in grayscale color map
+
+    """
+    plt.imshow(img.transpose((1, 2, 0))[:, :, 0], cmap='gray')
+
+
+def get_bbox_sz(bbox):
+    """
+    return the size of the bounding boxes
+    :param bbox: numpy.ndarray bounding box N by 4
+    :return: numpy.ndarray bounding box size, (width+length)/2
+    """
+    return (bbox[:, 2]+bbox[:, 3] - bbox[:, 0] - bbox[:, 1]) / 2
+