Method of rectifying text image, training method, electronic device, and medium

A method of rectifying a text image, the method comprising:

• performing, based on a gating strategy, a plurality of first layer-wise processing on a text image to be rectified, so as to obtain respective feature maps of a plurality of layer levels, wherein each of the feature maps comprises a text structural feature related to the text image to be rectified, and the gating strategy is configured to increase an attention to the text structural feature;
• and
• performing a plurality of second layer-wise processing on the respective feature maps of the plurality of layer levels, so as to obtain a rectified text image corresponding to the text image to be rectified,
• wherein the performing, based on a gating strategy, a plurality of first layer-wise processing comprises performing, based on a text image rectification model, a plurality of first layer-wise processing on the text image to be rectified, so as to obtain the respective feature maps of the plurality of layer levels, wherein the text image rectification model comprises a gating module created according to the gating strategy,
• wherein the text image rectification model further comprises an encoder and a decoder, and the gating module comprises a fine-grain layer unit;
• wherein the performing, based on a text image rectification model, a plurality of first layer-wise processing comprises performing, based on the encoder and the fine-grain layer unit, a plurality of first layer-wise processing on the text image to be rectified, so as to obtain the respective feature maps of the plurality of layer levels;
• and
• wherein the performing a plurality of second layer-wise processing on the respective feature maps comprises performing a plurality of second layer-wise processing on the respective feature maps of the plurality of layer levels based on the decoder, so as to obtain the rectified text image corresponding to the text image to be rectified,
• wherein the encoder comprises N down-sampling modules connected in cascade, and the decoder comprises N up-sampling modules connected in cascade, where N is an integer greater than 1;
• wherein the performing, based on the encoder and the fine-grain layer unit, a plurality of first layer-wise processing comprises:
• for 1<i≤N, processing a second down-sampling feature map of an (i−1)th layer level by using an ith down-sampling module, so as to obtain a second down-sampling feature map of an ith layer level;
• and
• processing the second down-sampling feature map of an Nth layer level by using the fine-grain layer unit, so as to obtain a second fine-grain feature map of the Nth layer level;
• and
• wherein the performing, based on the decoder, a plurality of second layer-wise processing comprises:
• for i=N, processing a fine-grain feature map of the Nth layer level by using an Nth up-sampling module, so as to obtain a second up-sampling feature map of the Nth layer level;
• fusing the second up-sampling feature map and the second down-sampling feature map of the Nth layer level to obtain a second fusion feature map of the Nth layer level;
• processing the second fusion feature map of the Nth layer level by using the Nth up-sampling module, so as to obtain a second output feature map of the Nth layer level;
• for 1≤i<N, processing a second output feature map of an (i+1)th layer level by using an ith up-sampling module, so as to obtain a second up-sampling feature map of the ith layer level;
• fusing the second down-sampling feature map and the second up-sampling feature map of the ith layer level to obtain a second fusion feature map of the ith layer level;
• processing the second fusion feature map of the ith layer level by using the ith up-sampling module, so as to obtain a second output feature map of the ith layer level;
• and
• determining, according to the second output feature map of a first layer level, the rectified text image corresponding to the text image to be rectified.

A method of rectifying a text image, the method comprising:

• performing, based on a gating strategy, a plurality of first layer-wise processing on a text image to be rectified, so as to obtain respective feature maps of a plurality of layer levels, wherein each of the feature maps comprises a text structural feature related to the text image to be rectified, and the gating strategy is configured to increase an attention to the text structural feature;
• and
• performing a plurality of second layer-wise processing on the respective feature maps of the plurality of layer levels, so as to obtain a rectified text image corresponding to the text image to be rectified,
• wherein the performing, based on a gating strategy, a plurality of first layer-wise processing comprises performing, based on a text image rectification model, a plurality of first layer-wise processing on the text image to be rectified, so as to obtain the respective feature maps of the plurality of layer levels, wherein the text image rectification model comprises a gating module created according to the gating strategy,
• wherein the text image rectification model further comprises an encoder and a decoder, and the gating module comprises a plurality of coarse-grain layer units;
• wherein the performing, based on a text image rectification model, a plurality of first layer-wise processing comprises performing, based on the encoder and the plurality of coarse-grain layer units, a plurality of first layer-wise processing on the text image to be rectified, so as to obtain the respective feature maps of the plurality of layer levels;
• and
• wherein the performing a plurality of second layer-wise processing comprises performing a plurality of second layer-wise processing on the respective feature maps of the plurality of layer levels based on the decoder, so as to obtain the rectified text image corresponding to the text image to be rectified,
• wherein the encoder comprises N down-sampling modules connected in cascade, the decoder comprises N up-sampling modules connected in cascade, and the gating module comprises N coarse-grain layer units, where N is an integer greater than 1;
• wherein the performing, based on the encoder and the plurality of coarse-grain layer units, a plurality of first layer-wise processing comprises:
• for 1<i≤N, processing a second down-sampling feature map of an (i−1)th layer level by using an ith down-sampling module, so as to obtain a second down-sampling feature map of the ith layer level;
• and
• processing the second down-sampling feature map of the ith layer level by using an ith coarse-grain layer unit, so as to obtain a second coarse-grain feature map of the ith layer level;
• and
• wherein the performing a plurality of second layer-wise processing on the respective feature maps of the plurality of layer levels based on the decoder comprises:
• for 1≤i<N, processing a third output feature map of an (i+1)th layer level by using an ith up-sampling module, so as to obtain a third up-sampling feature map of the ith layer level;
• fusing the second coarse-grain feature map and the third up-sampling feature map of the ith layer level to obtain a third fusion feature map of the ith layer level;
• processing the third fusion feature map of the ith layer level by using the ith up-sampling module, so as to obtain a third output feature map of the ith layer level;
• and
• determining, according to the third output feature map of a first layer level, the rectified text image corresponding to the text image to be rectified.