Automatic digital ECG signal extraction and normal QRS recognition from real scene ECG images.

•An Adaptive Filter Algorithm (AFA) is proposed in this paper for uneven illumination problem. The AFA method relieves the effect of uneven illumination that commonly exists in real-scene ECG image to a large extent.•A channel dependent hierarchical scheme for automatic ECG binary image extraction from ECG image in real scene is introduced. The two layers aim separately at gridline removing and illumination relieving depending on different channels’ properties, but they also complement each other and finally generate a good denoised ECG signal. 1D ECG signal extraction is based on ECG binary image and low difference between extracted value and real value shows excellent ability of the proposed approach.•We also bring forward a new QRS recognition method using fused image features. This provides a way to utilize spatial relationships of special points to assist in disease diagnose. A candidate point set strategy is used for further computing optimization.