添加对 POT 与 GNT 的数据处理 API

new file:   casia/image.py

Author: xinetzone

casia/image.py

@@ -0,0 +1,59 @@
+import struct
+import numpy as np
+
+
+class GNT:
+    '''GNT 文件的解码器'''
+    def __init__(self, Z, set_name):
+        self.Z = Z
+        self.set_name = set_name  # 数据集名称
+
+    def __iter__(self):
+        with self.Z.open(self.set_name) as fp:
+            head = True
+            while head:
+                head = fp.read(4)
+                if not head:  # 判断文件是否读到结尾
+                    break  # 读到文件结尾立即结束
+                head = struct.unpack('I', head)[0]
+                tag_code = fp.read(2).decode('gb18030')
+                width, height = struct.unpack('2H', fp.read(4))
+                bitmap = np.frombuffer(fp.read(width*height), np.uint8)
+                img = bitmap.reshape((height, width))
+                yield img, tag_code
+
+
+class POT:
+    '''POT 解码器'''
+    def __init__(self, Z, set_name):
+        self.Z = Z
+        self._fp = Z.open(set_name)
+
+    def __iter__(self):
+        size = struct.unpack('H', self._fp.read(2))[0]  # Sample size
+        tag = {}  # 记录字符与笔画
+        sizes = []
+        tag_id = 0
+        while size:
+            sizes.append(size)
+            tag_code = self._fp.read(4).decode(
+                'gb18030').strip('\x00')  # 字符解码
+            stroke_num = struct.unpack('H', self._fp.read(2))[0]  # 笔画数
+            strokes = {k: [] for k in range(stroke_num)}
+            k = 0
+            while k <= stroke_num:
+                xy = struct.unpack('2h', self._fp.read(4))
+                if xy == (-1, 0):
+                    k += 1
+                elif xy == (-1, -1):
+                    tag.update({tag_id: {tag_code: strokes}})  # 更新字典
+                    tag_id += 1
+                    size = self._fp.read(2)
+                    if size == b'':  # 判断是否解码完成
+                        ... # print('解码结束！')
+                    else:
+                        size = struct.unpack('H', size)[0]  # Sample size
+                    break
+                else:
+                    strokes[k].append(xy)  # 记录笔迹坐标
+            yield tag, sizes