<p>High-resolution ultrasonic phased array imaging is widely employed in industrial non-destructive testing and medical diagnostics, yet it faces two critical challenges: the massive computational load of full-matrix imaging algorithms and the “blind zone” caused by near-surface artifacts. To address these issues, this paper proposes a unified real-time imaging framework combining advanced signal processing with high-throughput computing. First, to mitigate the strong early reflection echoes generated by the transducer’s protective layer, a spatial-frequency filtering algorithm based on 2-D Discrete Cosine Transform (DCT) is proposed. This method effectively separates stationary reverberations from defect signals, significantly reducing near-field interference. Second, to overcome the computational bottleneck of the Total Focusing Method (TFM), a massively parallel beamforming strategy based on General-Purpose GPU (GPGPU) acceleration is implemented. Experimental results demonstrate that the proposed framework effectively eliminates film reverberations, yielding high-fidelity images with improved Contrast-to-Noise Ratio (CNR). Furthermore, the parallel implementation achieves real-time frame rates for high-resolution imaging tasks, validating its potential for practical engineering applications.</p>