Greetings to All,
I am working on camera captured document image analysis. i have studied some paper on features extraction from combination of two or more texture features. Now i wish to know what is the significance of the combining the many features to extract from document image. i have seen some papers where they have used features like (LBP+HoG), (LBP+LPQ+GLCM+HOG) and so on.
Please clarify things.
Thanks..
Dear sir,
You can use one feature extraction method such as DWT or SVD.
Please take a view to the following papers:
Article Low-Complexity Face Recognition using a Multilevel DWT and T...
Conference Paper A novel face recognition method based on one state of discre...
Conference Paper Using only two states of discrete HMM for high-speed face recognition
Best wishes
Dear Satish Kumar Mallappa,
All features you mentions is handcraft features which may not suitable with all data. Therefore the researcher usually performs the analysis using combination of those features.
If you wants to have a features extracted method that learnt directly from your data you can use deep CNN method.
Hopes it help.
Regards,
Combing many feature may not be efficient in terms of complexity. You should try to use deep CNN feature which can extract all hidden features from an image including edges, color, texture etc.
As per the above suggestions extracting the features by using deep CNN is suitable. But as per my knowledge to extract features from the images using deep CNN needs to be fixed dimension size of the image like 127x127, 227x227 and so on. But my images are not fix dimensions my dataset contains the word and line document images. which varying in size. so in this regard i need some support. Thank you.
Dear Satish Kumar Mallappa,
I don't know the goals of your problem. If you want to performs text detection or text recognition, you can try the state-of-the-art CNN method, such as YOLO or SSD.
Yes, you need to manage the input resolution to be fixed. You can either resize the input directly with/without preservation of aspect ratio or you can use a classifier that use convolutional operation in all layer.
Hopes it help.
Regards,
- Badges
- Science topic
- Similar topics
- Phytochemistry
- Extracts