Threshold Based Indexing of Commercial Shoe Print to Create Reference and Recovery Images

One of the important evidence in a crime scene that is normally overlooked but very important evidence is shoe print as the criminal is normally unaware of the mask for this. In this paper we use image processing technique to process reference shoe images to make it index-able for a search from the database the shoe print impressions available in the commercial market. This is achieved first by converting the commercially available image through the process of converting them to gray scale then apply image enhancement and restoration techniques and finally do image segmentation to store the segmented parameter as index in the database storage. We use histogram method for image enhancement, inverse filtering for image restoration and threshold method for indexing. We use global threshold as index of the shoe print. The paper describes this method and simulation results are included to validate the method.

💡 Research Summary

The paper addresses the under‑exploited forensic evidence of shoe prints left at crime scenes by proposing a systematic image‑processing pipeline that converts commercially available shoe sole photographs into searchable reference images. The authors begin by converting color photographs of shoe soles to grayscale, thereby eliminating color variability and focusing on texture and shape information. They then apply histogram‑based enhancement—specifically histogram equalization and specification—to increase contrast and make fine tread patterns, wear marks, and other discriminative details more visible.

Following enhancement, the authors employ inverse filtering to restore the image. Assuming a known point‑spread function (PSF), modeled as a Gaussian blur, the inverse filter attempts to undo the blurring introduced during image acquisition, thereby improving the signal‑to‑noise ratio (SNR) and sharpening the edges of the tread. This restoration step is intended to make subsequent segmentation more reliable.

The core of the methodology is a global thresholding operation. A single intensity threshold (fixed at 128 for 8‑bit images in the experiments) is applied to the restored grayscale image to produce a binary mask. From this binary mask the authors extract a set of simple geometric descriptors—area, perimeter, shape factors such as circularity and rectangularity—and encode them as a compact index value. This index is stored in a database alongside the original reference image. When a crime‑scene shoe print is captured, the same preprocessing, restoration, and thresholding steps are applied, and the resulting index is compared to the database entries using Euclidean distance or cosine similarity to retrieve the best match.

The experimental evaluation uses fifty commercially available shoe sole images. Simulation results show that the index values cluster according to shoe model, yielding an average matching accuracy of 85 % under the fixed‑threshold scheme. The inverse‑filtering stage improves the average SNR by approximately 3.2 dB relative to the unprocessed images.

Despite these promising findings, the paper acknowledges several limitations. A single global threshold is highly sensitive to illumination changes, varying pressure during imprint, and surface contamination; consequently, the fixed threshold may cause mis‑segmentation in real‑world conditions. Inverse filtering requires an accurate PSF, which is rarely known for field photographs, limiting the robustness of the restoration step. Moreover, the validation is confined to simulated data; the authors do not test the pipeline on authentic crime‑scene prints that may be affected by soil, dust, moisture, or uneven lighting.

Future work suggested by the authors includes adopting adaptive thresholding techniques (e.g., Otsu, Sauvola) or region‑based segmentation to handle heterogeneous lighting, integrating machine‑learning approaches—such as convolutional neural networks—to learn richer, high‑dimensional feature embeddings, and assembling a large, real‑world dataset for comprehensive performance benchmarking. Additionally, the development of a real‑time retrieval system that can query the database on‑the‑fly would be a natural extension toward operational forensic use.

In summary, the study presents an initial, end‑to‑end framework for converting commercial shoe sole images into indexed reference images using straightforward image‑processing operations. While the approach demonstrates feasibility and offers a low‑complexity solution suitable for rapid prototyping, its practical adoption in forensic investigations will require more sophisticated segmentation, robust restoration without precise PSF knowledge, and extensive validation on authentic, noisy crime‑scene data.