Azure Cosmos DB is not intended to be used to store images. To do so means encoding as base-64. This can be incredibly expensive compute-wise to do. Also, max doc size for Cosmos is 2MB. Anything larger would need to be shredded and reassembled client-side. Lastly, storage for Cosmos DB is more expensive than blob storage, 25c/GB versus 1c/GB for blob storage. Using Cosmos to store b64 encoded image data is expensive given you cannot index or query the image data itself.
The right approach is to put the image in blob-storage, then include the URI to the image and all searchable metadata in Cosmos. If you're building a search engine you can roll your own or use Azure Cognitive Search which has very good integration with Cosmos DB, https://learn.microsoft.com/en-us/azure/search/search-howto-index-cosmosdb